[slurm-dev] Re: fairshare
Bill, I may be wrong (corrections welcomed), but I'm pretty sure you'll have to use a database query. My understanding is that the decayed usage is stored as a single usage_raw value per association (https://github.com/SchedMD/slurm/blob/f8025c1484838ecbe3e690fa565452d990123361/src/plugins/priority/multifactor/priority_multifactor.c#L1119). There is no history of any kind. You would have to do a fairly complex query to get an accurate representation or write some code to recreate the way Slurm does it. If you look at _apply_decay() and _apply_new_usage() in src/plugins/priority/multifactor/priority_multifactor.c, you can see all that happens. Basically, once per decay thread iteration each association's usage_raw and the job's cputime for that time period is calculated and decayed accordingly. This can happen many, many times over the length of a job. If a job terminates before reaching its timelimit, the remaining allocated cputime is immediately added all at the same time (https://github.com/SchedMD/slurm/blob/f8025c1484838ecbe3e690fa565452d990123361/src/plugins/priority/multifactor/priority_multifactor.c#L1036). Those are some of the issues that you may run into while creating a database tool for this. I could be mistaken on some of the details but that is my understanding of the code (we looked recently for an unrelated reason). Ryan On 07/14/2014 02:15 PM, Bill Wichser wrote: Is there any way to get a better view of fairshare than the "sshare" command? Under PBS, there was the diagnose -f command which showed the breakdown per set time period which calculated this value. What was nice about this was I could point a group to this command, or cut and paste, showing that you have been using 20% over the last 30 days even though you haven't run anything in the last three days. It's a much more difficult problem when asked now. I have no tool which shows the value, and decay, over the time. So I'm wondering if anyone has a method to demonstrate that, yes, this fairshare value is correct and here is why. Or do I just need to figure out a database query to cull this information? Thanks, Bill -- Ryan Cox Operations Director Fulton Supercomputing Lab Brigham Young University
[slurm-dev] Re: fairshare usage
Have you looked at sshare? Phil Eckert LLNL From: Mario Kadastik mailto:mario.kadas...@cern.ch>> Reply-To: slurm-dev mailto:slurm-dev@schedmd.com>> Date: Tuesday, January 22, 2013 11:17 AM To: slurm-dev mailto:slurm-dev@schedmd.com>> Subject: [slurm-dev] fairshare usage Hi, is there some decent way to get multifactor fairshare current state? Something akin to maui's diagnose -f output that shows groups (accounts for slurm) and users with their fairshare target as well as their historic usage over the past N days. This would seriously help understand how the fairshare is computed based on the actual usage statistics and current cluster state. For example we have all user fairshares set as parent and for the accounts: Account Share -- - root 1 grid 1 grid-ops 1 hepusers 100 kbfiusers 1 now let's assume one of the users in hepusers spends the past N days computing with the full cluster and then another user submits a number of jobs it would be logical to assume that as there is no distinction between the users in an account the newcomers priority would be higher as (s)he hasn't had any allocated time. [root@slurm-1 slurm]# sreport cluster accountutilizationbyuser start=2013-01-08 Cluster/Account/User Utilization 2013-01-08T00:00:00 - 2013-01-21T23:59:59 (1209600 secs) Time reported in CPU Minutes Cluster Account Login Proper Name Used - --- - --- -- t2estoniaroot 7801048 t2estoniagrid0 t2estoniagridcms134 mapped user fo+ 0 t2estoniagrid sgmcms000 mapped user fo+ 0 t2estoniahepusers 7801048 t2estoniahepusersandres Andres Tiko 85048 t2estoniahepusers mario Mario Kadastik7716000 so according to this Mario (me) has computed a huge amount of time in comparison to andres. However if I look at the priorities from sinfo -nl I see this: [root@slurm-1 slurm]# sprio -nl|head -3 JOBID USER PRIORITY AGEFAIRSHARE JOBSIZEPARTITION QOS 53498mario 0.3497 0.2404977 0.4897101 0.9919238 1.000 0.000 53499mario 0.3497 0.2404977 0.4897101 0.9919238 1.000 0.000 [root@slurm-1 slurm]# sprio -nl|grep andres|head -1 53835 andres 0.3497 0.2396412 0.4897101 0.9919238 1.000 0.000 so in fact the fairshare factor is equivalent for both users no matter that one has been getting a lot of the resource while the other has not. or do I misunderstand the =parent part? I tried also setting all users shares to 1 and have no clue how long it will take for sprio to recompute this, but right now it's showing the same priorities. That's one of the reasons why I'd like to be able to see how the actual usage and decay over time affect the factor so that I can better understand the algorithm and tune the weights. Thanks, Mario Kadastik, PhD Researcher --- "Physics is like sex, sure it may have practical reasons, but that's not why we do it" -- Richard P. Feynman
[slurm-dev] Re: fairshare usage
Look at the sshare and sprio tools. Quoting Mario Kadastik : > Hi, > > is there some decent way to get multifactor fairshare current state? > Something akin to maui's diagnose -f output that shows groups > (accounts for slurm) and users with their fairshare target as well > as their historic usage over the past N days. This would seriously > help understand how the fairshare is computed based on the actual > usage statistics and current cluster state. > > For example we have all user fairshares set as parent and for the accounts: > >Account Share > -- - > root 1 > grid 1 > grid-ops 1 > hepusers 100 > kbfiusers 1 > > now let's assume one of the users in hepusers spends the past N days > computing with the full cluster and then another user submits a > number of jobs it would be logical to assume that as there is no > distinction between the users in an account the newcomers priority > would be higher as (s)he hasn't had any allocated time. > > [root@slurm-1 slurm]# sreport cluster accountutilizationbyuser > start=2013-01-08 > > Cluster/Account/User Utilization 2013-01-08T00:00:00 - > 2013-01-21T23:59:59 (1209600 secs) > Time reported in CPU Minutes > > Cluster Account Login Proper Name Used > - --- - --- -- > t2estoniaroot 7801048 > t2estoniagrid0 > t2estoniagridcms134 mapped user fo+ 0 > t2estoniagrid sgmcms000 mapped user fo+ 0 > t2estoniahepusers 7801048 > t2estoniahepusersandres Andres Tiko 85048 > t2estoniahepusers mario Mario Kadastik7716000 > > so according to this Mario (me) has computed a huge amount of time > in comparison to andres. However if I look at the priorities from > sinfo -nl I see this: > > [root@slurm-1 slurm]# sprio -nl|head -3 > JOBID USER PRIORITY AGEFAIRSHARE JOBSIZEPARTITION QOS > 53498mario 0.3497 0.2404977 0.4897101 0.9919238 > 1.000 0.000 > 53499mario 0.3497 0.2404977 0.4897101 0.9919238 > 1.000 0.000 > [root@slurm-1 slurm]# sprio -nl|grep andres|head -1 > 53835 andres 0.3497 0.2396412 0.4897101 0.9919238 > 1.000 0.000 > > so in fact the fairshare factor is equivalent for both users no > matter that one has been getting a lot of the resource while the > other has not. > > or do I misunderstand the =parent part? I tried also setting all > users shares to 1 and have no clue how long it will take for sprio > to recompute this, but right now it's showing the same priorities. > > That's one of the reasons why I'd like to be able to see how the > actual usage and decay over time affect the factor so that I can > better understand the algorithm and tune the weights. > > Thanks, > > Mario Kadastik, PhD > Researcher > > --- > "Physics is like sex, sure it may have practical reasons, but > that's not why we do it" > -- Richard P. Feynman > >
[slurm-dev] Re: fairshare incrementing
We think we may be onto something, in sacct we were looking at the jobs submitted by the users, and found that many users share the same uidnumber in the slurm database. It seems to correlate with the size of the user's uid number in our ldap directory... users who's uid number are greater than 65535 get trunked to that number... users with uid numbers below that keep their correct uidnumbers (user2 in the sample output below) [root@slurm-master ~]# sacct -c --format=User,uid,JobID,JobName,NodeList,Start,Elapsed,ExitCode,DerivedExitCode,state |grep user2|head user2 27545 30548 bwa node01-1 2013-07-08T13:04:25 00:00:48 0:0 COMPLETED user2 27545 30571 bwa node01-1 2013-07-08T15:18:00 00:00:48 0:0 COMPLETED user2 27545 30573 bwa node01-1 2013-07-09T09:40:59 00:00:48 0:0 COMPLETED user2 27545 30618 grep node01-1 2013-07-09T11:57:12 00:00:48 0:0 COMPLETED user2 27545 30619bc node01-1 2013-07-09T11:58:08 00:00:48 0:0 CANCELLED user2 27545 30620du node01-1 2013-07-09T11:58:19 00:00:48 0:0 COMPLETED user2 27545 30621wc node01-1 2013-07-09T11:58:43 00:00:48 0:0 COMPLETED user2 27545 30622 zcat node01-1 2013-07-09T11:58:54 00:00:48 0:0 COMPLETED user2 27545 30623 zcat node01-1 2013-07-09T12:12:56 00:00:48 0:0 COMPLETED user2 27545 30624 zcat node01-1 2013-07-09T12:26:37 00:00:48 0:0 CANCELLED [root@slurm-master ~]# sacct -c --format=User,uid,JobID,JobName,NodeList,Start,Elapsed,ExitCode,DerivedExitCode,state |grep user1|head user1 65535 83 impute2_w+ node01-1 2013-04-17T09:29:47 00:00:48 0:0 FAILED user1 65535 84 impute2_w+ node01-1 2013-04-17T09:30:17 00:00:48 0:0 FAILED user1 65535 85 impute2_w+ node01-1 2013-04-17T09:30:40 00:00:48 0:0 FAILED user1 65535 86 impute2_w+ node01-1 2013-04-17T09:40:45 00:00:48 0:0 FAILED user1 65535 87 date node01-1 2013-04-17T09:42:36 00:00:48 0:0 COMPLETED user1 65535 88 hostname node01-1 2013-04-17T09:42:37 00:00:48 0:0 COMPLETED user1 65535 89 impute2_w+ node01-1 2013-04-17T09:48:50 00:00:48 0:0 FAILED user1 65535 90 impute2_w+ node01-1 2013-04-17T09:48:56 00:00:48 0:0 FAILED user1 65535 91 impute2_w+ node01-1 2013-04-17T09:49:56 00:00:48 0:0 FAILED user1 65535 92 impute2_w+ node01-1 2013-04-17T09:50:06 00:00:48 0:0 FAILED [root@slurm-master ~]# sacct -c --format=User,uid,JobID,JobName,NodeList,Start,Elapsed,ExitCode,DerivedExitCode,state |grep user3|head user3 65535 5 script.sh node09-1 2013-04-09T15:55:07 00:00:48 0:0 FAILED user3 65535 6 script.sh node09-1 2013-04-09T15:55:13INVALID 0:0 COMPLETED user3 65535 8 bash node09-1 2013-04-09T15:57:34 00:00:48 0:0 COMPLETED user3 65535 7 bash node09-1 2013-04-09T15:57:21 00:00:48 0:0 COMPLETED user3 65535 23script.sh node09-1 2013-04-09T16:10:02 00:00:48 0:0 COMPLETED user3 65535 27script.sh node09-+ 2013-04-09T16:18:33 00:00:48 0:0 CANCELLED user3 65535 28script.sh node01-+ 2013-04-09T16:18:55 00:00:48 0:0 CANCELLED user3 65535 30script.sh node01-+ 2013-04-09T16:34:12 00:00:48 0:0 CANCELLED user3 65535 31script.sh node01-+ 2013-04-09T16:34:17 00:00:48 0:0 CANCELLED user3 65535 32script.sh node01-+ 2013-04-09T16:34:21 00:00:48 0:0 CANCELLED We are thinking perhaps this could lead to our major issues with the system and priority factoring. AC On 08/23/2013 07:56 AM, Alan V. Cowles wrote: Hey guys, So in the past we had 3 prioritization factors in effect: partition, age and fairshare and they were working wonderfully. Currently partition has no effect for us as it's all one large shared partition so everyone gets the same value there. So everything is balanced in age and fairshare, In the past age and fairshare worked splendidly, and we have it set as I understand to refresh counters every 2 weeks... so basically everyone had a blank slate this past weekend. What our current issue is as follows... A problematic user has submitted 70k job
[slurm-dev] Re: fairshare incrementing
Sorry to spam the list, but we wanted to keep updates in flux. We managed to find the issue in our mysqldb we are using for job accounting which had the column value set to smallint (5) for that value, so it was rounding things off, some SQL magic and we now have appropriate uid's showing up. A new monkey wrench, some test jobs submitted by user3 below get their fairshare value of 5000 as expected, just not user2... we just cleared his jobs from the queue, and submitted another 100 jobs for testing and none of them got a fairshare value... In his entire history of using our cluster he hasn't submitted over 5000 jobs, in fact: [root@slurm-master ~]# sacct -c --format=user,jobid,jobname,start,elapsed,state,exitcode -u user2 | grep user2 | wc -l 2573 So we can't figure out why he's being overlooked. AC On 08/23/2013 10:31 AM, Alan V. Cowles wrote: We think we may be onto something, in sacct we were looking at the jobs submitted by the users, and found that many users share the same uidnumber in the slurm database. It seems to correlate with the size of the user's uid number in our ldap directory... users who's uid number are greater than 65535 get trunked to that number... users with uid numbers below that keep their correct uidnumbers (user2 in the sample output below) [root@slurm-master ~]# sacct -c --format=User,uid,JobID,JobName,NodeList,Start,Elapsed,ExitCode,DerivedExitCode,state |grep user2|head user2 27545 30548 bwa node01-1 2013-07-08T13:04:25 00:00:48 0:0 COMPLETED user2 27545 30571 bwa node01-1 2013-07-08T15:18:00 00:00:48 0:0 COMPLETED user2 27545 30573 bwa node01-1 2013-07-09T09:40:59 00:00:48 0:0 COMPLETED user2 27545 30618 grep node01-1 2013-07-09T11:57:12 00:00:48 0:0 COMPLETED user2 27545 30619bc node01-1 2013-07-09T11:58:08 00:00:48 0:0 CANCELLED user2 27545 30620du node01-1 2013-07-09T11:58:19 00:00:48 0:0 COMPLETED user2 27545 30621wc node01-1 2013-07-09T11:58:43 00:00:48 0:0 COMPLETED user2 27545 30622 zcat node01-1 2013-07-09T11:58:54 00:00:48 0:0 COMPLETED user2 27545 30623 zcat node01-1 2013-07-09T12:12:56 00:00:48 0:0 COMPLETED user2 27545 30624 zcat node01-1 2013-07-09T12:26:37 00:00:48 0:0 CANCELLED [root@slurm-master ~]# sacct -c --format=User,uid,JobID,JobName,NodeList,Start,Elapsed,ExitCode,DerivedExitCode,state |grep user1|head user1 65535 83 impute2_w+ node01-1 2013-04-17T09:29:47 00:00:48 0:0 FAILED user1 65535 84 impute2_w+ node01-1 2013-04-17T09:30:17 00:00:48 0:0 FAILED user1 65535 85 impute2_w+ node01-1 2013-04-17T09:30:40 00:00:48 0:0 FAILED user1 65535 86 impute2_w+ node01-1 2013-04-17T09:40:45 00:00:48 0:0 FAILED user1 65535 87 date node01-1 2013-04-17T09:42:36 00:00:48 0:0 COMPLETED user1 65535 88 hostname node01-1 2013-04-17T09:42:37 00:00:48 0:0 COMPLETED user1 65535 89 impute2_w+ node01-1 2013-04-17T09:48:50 00:00:48 0:0 FAILED user1 65535 90 impute2_w+ node01-1 2013-04-17T09:48:56 00:00:48 0:0 FAILED user1 65535 91 impute2_w+ node01-1 2013-04-17T09:49:56 00:00:48 0:0 FAILED user1 65535 92 impute2_w+ node01-1 2013-04-17T09:50:06 00:00:48 0:0 FAILED [root@slurm-master ~]# sacct -c --format=User,uid,JobID,JobName,NodeList,Start,Elapsed,ExitCode,DerivedExitCode,state |grep user3|head user3 65535 5 script.sh node09-1 2013-04-09T15:55:07 00:00:48 0:0 FAILED user3 65535 6 script.sh node09-1 2013-04-09T15:55:13 INVALID 0:0 COMPLETED user3 65535 8 bash node09-1 2013-04-09T15:57:34 00:00:48 0:0 COMPLETED user3 65535 7 bash node09-1 2013-04-09T15:57:21 00:00:48 0:0 COMPLETED user3 65535 23script.sh node09-1 2013-04-09T16:10:02 00:00:48 0:0 COMPLETED user3 65535 27script.sh node09-+ 2013-04-09T16:18:33 00:00:48 0:0 CANCELLED user3 65535 28script.sh node01-+ 2013-04-09T16:18:55 00:00:48 0:0 CANCELLED user3 65535 30script.sh node01-+ 2013-04-09T16:34:12 00:00:48 0:0 CANCELLED user3 65535 31script.sh node01-+ 2013-04-09T16:34:17 00:00:4
[slurm-dev] Re: fairshare incrementing
Final update for the day, we have found what is causing priority to be overlooked we just don't know what is causing it... [root@cluster-login ~]# squeue --format="%a %.7i %.9P %.8j %.8u %.8T %.10M %.9l %.6D %R" |grep user1 (null) 181378lowmem testbatc user1 PENDING 0:00 UNLIMITED 1 (Priority) (null) 181379lowmem testbatc user1 PENDING 0:00 UNLIMITED 1 (Priority) (null) 181380lowmem testbatc user1 PENDING 0:00 UNLIMITED 1 (Priority) (null) 181381lowmem testbatc user1 PENDING 0:00 UNLIMITED 1 (Priority) (null) 181382lowmem testbatc user1 PENDING 0:00 UNLIMITED 1 (Priority) (null) 181383lowmem testbatc user1 PENDING 0:00 UNLIMITED 1 (Priority) (null) 181384lowmem testbatc user1 PENDING 0:00 UNLIMITED 1 (Priority) (null) 181385lowmem testbatc user1 PENDING 0:00 UNLIMITED 1 (Priority) (null) 181386lowmem testbatc user1 PENDING 0:00 UNLIMITED 1 (Priority) (null) 181387lowmem testbatc user1 PENDING 0:00 UNLIMITED 1 (Priority) Compared to: [root@cluster-login ~]# squeue --format="%a %.7i %.9P %.8j %.8u %.8T %.10M %.9l %.6D %R" |grep user2 account 181378lowmem testbatc user2 PENDING 0:00 UNLIMITED 1 (Priority) account 181379lowmem testbatc user2 PENDING 0:00 UNLIMITED 1 (Priority) account 181380lowmem testbatc user2 PENDING 0:00 UNLIMITED 1 (Priority) account 181381lowmem testbatc user2 PENDING 0:00 UNLIMITED 1 (Priority) account 181382lowmem testbatc user2 PENDING 0:00 UNLIMITED 1 (Priority) account 181383lowmem testbatc user2 PENDING 0:00 UNLIMITED 1 (Priority) account 181384lowmem testbatc user2 PENDING 0:00 UNLIMITED 1 (Priority) account 181385lowmem testbatc user2 PENDING 0:00 UNLIMITED 1 (Priority) account 181386lowmem testbatc user2 PENDING 0:00 UNLIMITED 1 (Priority) account 181387lowmem testbatc user2 PENDING 0:00 UNLIMITED 1 (Priority) We have tried to create new users and new accounts this afternoon and all of them show (null) as their account when we break out the formatting rules on sacct. sacctmgr add account accountname sacctmgr add user username defaultaccount accountname We have even one case where all users under and account are working fine except a user we added yesterday... so at some point in the past (logs aren't helping us thus far) the ability to actually sync up a user and an account for accounting purposes has left us. Also I have failed to mention to this point that we are still running Slurm 2.5.4, my apologies for that. AC On 08/23/2013 11:22 AM, Alan V. Cowles wrote: Sorry to spam the list, but we wanted to keep updates in flux. We managed to find the issue in our mysqldb we are using for job accounting which had the column value set to smallint (5) for that value, so it was rounding things off, some SQL magic and we now have appropriate uid's showing up. A new monkey wrench, some test jobs submitted by user3 below get their fairshare value of 5000 as expected, just not user2... we just cleared his jobs from the queue, and submitted another 100 jobs for testing and none of them got a fairshare value... In his entire history of using our cluster he hasn't submitted over 5000 jobs, in fact: [root@slurm-master ~]# sacct -c --format=user,jobid,jobname,start,elapsed,state,exitcode -u user2 | grep user2 | wc -l 2573 So we can't figure out why he's being overlooked. AC On 08/23/2013 10:31 AM, Alan V. Cowles wrote: We think we may be onto something, in sacct we were looking at the jobs submitted by the users, and found that many users share the same uidnumber in the slurm database. It seems to correlate with the size of the user's uid number in our ldap directory... users who's uid number are greater than 65535 get trunked to that number... users with uid numbers below that keep their correct uidnumbers (user2 in the sample output below) [root@slurm-master ~]# sacct -c --format=User,uid,JobID,JobName,NodeList,Start,Elapsed,ExitCode,DerivedExitCode,state |grep user2|head user2 27545 30548 bwa node01-1 2013-07-08T13:04:25 00:00:48 0:0 COMPLETED user2 27545 30571 bwa node01-1 2013-07-08T15:18:00 00:00:48 0:0 COMPLETED user2 27545 30573 bwa node01-1 2013-07-09T09:40:59 00:00:48 0:0 COMPLETED user2 27545 30618 grep node01-1 2013-07-09T11:57:12 00:00:48 0:0 COMPLETED user2 27545 30619bc node01-1 2013-07-09T11:58:08 00:00:48 0:0 CANCELLED user2 27545 30620du node01-1 2013-07-09T11:58:19 00:00:48 0:0 COMPLETED user2 27545 30621wc node01-1 2013-07-09T11:58:43 00:00:48 0:0 COMPLETED user2 27545 30622 zcat node01-1 2013-07-09T11:58:54 00:00:48
[slurm-dev] Re: fairshare incrementing
Perhaps it is a copy/paste error - but those two tables are identical On Aug 23, 2013, at 12:14 PM, Alan V. Cowles wrote: > > Final update for the day, we have found what is causing priority to be > overlooked we just don't know what is causing it... > > [root@cluster-login ~]# squeue --format="%a %.7i %.9P %.8j %.8u %.8T %.10M > %.9l %.6D %R" |grep user1 > (null) 181378lowmem testbatc user1 PENDING 0:00 UNLIMITED1 > (Priority) > (null) 181379lowmem testbatc user1 PENDING 0:00 UNLIMITED1 > (Priority) > (null) 181380lowmem testbatc user1 PENDING 0:00 UNLIMITED1 > (Priority) > (null) 181381lowmem testbatc user1 PENDING 0:00 UNLIMITED1 > (Priority) > (null) 181382lowmem testbatc user1 PENDING 0:00 UNLIMITED1 > (Priority) > (null) 181383lowmem testbatc user1 PENDING 0:00 UNLIMITED1 > (Priority) > (null) 181384lowmem testbatc user1 PENDING 0:00 UNLIMITED1 > (Priority) > (null) 181385lowmem testbatc user1 PENDING 0:00 UNLIMITED1 > (Priority) > (null) 181386lowmem testbatc user1 PENDING 0:00 UNLIMITED1 > (Priority) > (null) 181387lowmem testbatc user1 PENDING 0:00 UNLIMITED1 > (Priority) > > Compared to: > > [root@cluster-login ~]# squeue --format="%a %.7i %.9P %.8j %.8u %.8T %.10M > %.9l %.6D %R" |grep user2 > account 181378lowmem testbatc user2 PENDING 0:00 UNLIMITED1 > (Priority) > account 181379lowmem testbatc user2 PENDING 0:00 UNLIMITED1 > (Priority) > account 181380lowmem testbatc user2 PENDING 0:00 UNLIMITED1 > (Priority) > account 181381lowmem testbatc user2 PENDING 0:00 UNLIMITED1 > (Priority) > account 181382lowmem testbatc user2 PENDING 0:00 UNLIMITED1 > (Priority) > account 181383lowmem testbatc user2 PENDING 0:00 UNLIMITED1 > (Priority) > account 181384lowmem testbatc user2 PENDING 0:00 UNLIMITED1 > (Priority) > account 181385lowmem testbatc user2 PENDING 0:00 UNLIMITED1 > (Priority) > account 181386lowmem testbatc user2 PENDING 0:00 UNLIMITED1 > (Priority) > account 181387lowmem testbatc user2 PENDING 0:00 UNLIMITED1 > (Priority) > > > We have tried to create new users and new accounts this afternoon and all of > them show (null) as their account when we break out the formatting rules on > sacct. > > sacctmgr add account accountname > sacctmgr add user username defaultaccount accountname > > We have even one case where all users under and account are working fine > except a user we added yesterday... so at some point in the past (logs aren't > helping us thus far) the ability to actually sync up a user and an account > for accounting purposes has left us. Also I have failed to mention to this > point that we are still running Slurm 2.5.4, my apologies for that. > > AC > > > On 08/23/2013 11:22 AM, Alan V. Cowles wrote: >> Sorry to spam the list, but we wanted to keep updates in flux. >> >> We managed to find the issue in our mysqldb we are using for job accounting >> which had the column value set to smallint (5) for that value, so it was >> rounding things off, some SQL magic and we now have appropriate uid's >> showing up. A new monkey wrench, some test jobs submitted by user3 below get >> their fairshare value of 5000 as expected, just not user2... we just cleared >> his jobs from the queue, and submitted another 100 jobs for testing and none >> of them got a fairshare value... >> >> In his entire history of using our cluster he hasn't submitted over 5000 >> jobs, in fact: >> >> [root@slurm-master ~]# sacct -c >> --format=user,jobid,jobname,start,elapsed,state,exitcode -u user2 | grep >> user2 | wc -l >> 2573 >> >> So we can't figure out why he's being overlooked. >> >> AC >> >> >> On 08/23/2013 10:31 AM, Alan V. Cowles wrote: >>> We think we may be onto something, in sacct we were looking at the jobs >>> submitted by the users, and found that many users share the same uidnumber >>> in the slurm database. It seems to correlate with the size of the user's >>> uid number in our ldap directory... users who's uid number are greater than >>> 65535 get trunked to that number... users with uid numbers below that keep >>> their correct uidnumbers (user2 in the sample output below) >>> >>> >>> >>> >>> [root@slurm-master ~]# sacct -c >>> --format=User,uid,JobID,JobName,NodeList,Start,Elapsed,ExitCode,DerivedExitCode,state >>> |grep user2|head >>> user2 27545 30548 bwa node01-1 2013-07-08T13:04:25 >>> 00:00:48 0:0 COMPLETED >>> user2 27545 30571 bwa node01-1 2013-07-08T15:18:00 >>> 00:00:48 0:0 COMPLETED >>> user2 27545 30573 bwa node01-1 2013-07-09T09:40:59 >>> 00:00:48 0:0 COMPLETED >>> user2 27545 30618 grep node01-1 2013-07-09T11:57:12
[slurm-dev] Re: fairshare incrementing
Ah, never mind - I see the difference now. Was looking for some info to be different On Aug 23, 2013, at 12:17 PM, Ralph Castain wrote: > Perhaps it is a copy/paste error - but those two tables are identical > > On Aug 23, 2013, at 12:14 PM, Alan V. Cowles wrote: > >> >> Final update for the day, we have found what is causing priority to be >> overlooked we just don't know what is causing it... >> >> [root@cluster-login ~]# squeue --format="%a %.7i %.9P %.8j %.8u %.8T %.10M >> %.9l %.6D %R" |grep user1 >> (null) 181378lowmem testbatc user1 PENDING 0:00 UNLIMITED1 >> (Priority) >> (null) 181379lowmem testbatc user1 PENDING 0:00 UNLIMITED1 >> (Priority) >> (null) 181380lowmem testbatc user1 PENDING 0:00 UNLIMITED1 >> (Priority) >> (null) 181381lowmem testbatc user1 PENDING 0:00 UNLIMITED1 >> (Priority) >> (null) 181382lowmem testbatc user1 PENDING 0:00 UNLIMITED1 >> (Priority) >> (null) 181383lowmem testbatc user1 PENDING 0:00 UNLIMITED1 >> (Priority) >> (null) 181384lowmem testbatc user1 PENDING 0:00 UNLIMITED1 >> (Priority) >> (null) 181385lowmem testbatc user1 PENDING 0:00 UNLIMITED1 >> (Priority) >> (null) 181386lowmem testbatc user1 PENDING 0:00 UNLIMITED1 >> (Priority) >> (null) 181387lowmem testbatc user1 PENDING 0:00 UNLIMITED1 >> (Priority) >> >> Compared to: >> >> [root@cluster-login ~]# squeue --format="%a %.7i %.9P %.8j %.8u %.8T %.10M >> %.9l %.6D %R" |grep user2 >> account 181378lowmem testbatc user2 PENDING 0:00 UNLIMITED1 >> (Priority) >> account 181379lowmem testbatc user2 PENDING 0:00 UNLIMITED1 >> (Priority) >> account 181380lowmem testbatc user2 PENDING 0:00 UNLIMITED1 >> (Priority) >> account 181381lowmem testbatc user2 PENDING 0:00 UNLIMITED1 >> (Priority) >> account 181382lowmem testbatc user2 PENDING 0:00 UNLIMITED1 >> (Priority) >> account 181383lowmem testbatc user2 PENDING 0:00 UNLIMITED1 >> (Priority) >> account 181384lowmem testbatc user2 PENDING 0:00 UNLIMITED1 >> (Priority) >> account 181385lowmem testbatc user2 PENDING 0:00 UNLIMITED1 >> (Priority) >> account 181386lowmem testbatc user2 PENDING 0:00 UNLIMITED1 >> (Priority) >> account 181387lowmem testbatc user2 PENDING 0:00 UNLIMITED1 >> (Priority) >> >> >> We have tried to create new users and new accounts this afternoon and all of >> them show (null) as their account when we break out the formatting rules on >> sacct. >> >> sacctmgr add account accountname >> sacctmgr add user username defaultaccount accountname >> >> We have even one case where all users under and account are working fine >> except a user we added yesterday... so at some point in the past (logs >> aren't helping us thus far) the ability to actually sync up a user and an >> account for accounting purposes has left us. Also I have failed to mention >> to this point that we are still running Slurm 2.5.4, my apologies for that. >> >> AC >> >> >> On 08/23/2013 11:22 AM, Alan V. Cowles wrote: >>> Sorry to spam the list, but we wanted to keep updates in flux. >>> >>> We managed to find the issue in our mysqldb we are using for job accounting >>> which had the column value set to smallint (5) for that value, so it was >>> rounding things off, some SQL magic and we now have appropriate uid's >>> showing up. A new monkey wrench, some test jobs submitted by user3 below >>> get their fairshare value of 5000 as expected, just not user2... we just >>> cleared his jobs from the queue, and submitted another 100 jobs for testing >>> and none of them got a fairshare value... >>> >>> In his entire history of using our cluster he hasn't submitted over 5000 >>> jobs, in fact: >>> >>> [root@slurm-master ~]# sacct -c >>> --format=user,jobid,jobname,start,elapsed,state,exitcode -u user2 | grep >>> user2 | wc -l >>> 2573 >>> >>> So we can't figure out why he's being overlooked. >>> >>> AC >>> >>> >>> On 08/23/2013 10:31 AM, Alan V. Cowles wrote: We think we may be onto something, in sacct we were looking at the jobs submitted by the users, and found that many users share the same uidnumber in the slurm database. It seems to correlate with the size of the user's uid number in our ldap directory... users who's uid number are greater than 65535 get trunked to that number... users with uid numbers below that keep their correct uidnumbers (user2 in the sample output below) [root@slurm-master ~]# sacct -c --format=User,uid,JobID,JobName,NodeList,Start,Elapsed,ExitCode,DerivedExitCode,state |grep user2|head user2 27545 30548 bwa node01-1 2013-07-08T13:04:25 00:00:48 0:0 COMPLETED user2 27545 305
[slurm-dev] Re: fairshare incrementing
Yes its all us just running and killing test jobs as the various users now. Ralph Castain wrote: > >Ah, never mind - I see the difference now. Was looking for some info to be >different > > >On Aug 23, 2013, at 12:17 PM, Ralph Castain wrote: > >> Perhaps it is a copy/paste error - but those two tables are identical >> >> On Aug 23, 2013, at 12:14 PM, Alan V. Cowles wrote: >> >>> >>> Final update for the day, we have found what is causing priority to be >>> overlooked we just don't know what is causing it... >>> >>> [root@cluster-login ~]# squeue --format="%a %.7i %.9P %.8j %.8u %.8T %.10M >>> %.9l %.6D %R" |grep user1 >>> (null) 181378lowmem testbatc user1 PENDING 0:00 UNLIMITED1 >>> (Priority) >>> (null) 181379lowmem testbatc user1 PENDING 0:00 UNLIMITED1 >>> (Priority) >>> (null) 181380lowmem testbatc user1 PENDING 0:00 UNLIMITED1 >>> (Priority) >>> (null) 181381lowmem testbatc user1 PENDING 0:00 UNLIMITED1 >>> (Priority) >>> (null) 181382lowmem testbatc user1 PENDING 0:00 UNLIMITED1 >>> (Priority) >>> (null) 181383lowmem testbatc user1 PENDING 0:00 UNLIMITED1 >>> (Priority) >>> (null) 181384lowmem testbatc user1 PENDING 0:00 UNLIMITED1 >>> (Priority) >>> (null) 181385lowmem testbatc user1 PENDING 0:00 UNLIMITED1 >>> (Priority) >>> (null) 181386lowmem testbatc user1 PENDING 0:00 UNLIMITED1 >>> (Priority) >>> (null) 181387lowmem testbatc user1 PENDING 0:00 UNLIMITED1 >>> (Priority) >>> >>> Compared to: >>> >>> [root@cluster-login ~]# squeue --format="%a %.7i %.9P %.8j %.8u %.8T %.10M >>> %.9l %.6D %R" |grep user2 >>> account 181378lowmem testbatc user2 PENDING 0:00 UNLIMITED1 >>> (Priority) >>> account 181379lowmem testbatc user2 PENDING 0:00 UNLIMITED1 >>> (Priority) >>> account 181380lowmem testbatc user2 PENDING 0:00 UNLIMITED1 >>> (Priority) >>> account 181381lowmem testbatc user2 PENDING 0:00 UNLIMITED1 >>> (Priority) >>> account 181382lowmem testbatc user2 PENDING 0:00 UNLIMITED1 >>> (Priority) >>> account 181383lowmem testbatc user2 PENDING 0:00 UNLIMITED1 >>> (Priority) >>> account 181384lowmem testbatc user2 PENDING 0:00 UNLIMITED1 >>> (Priority) >>> account 181385lowmem testbatc user2 PENDING 0:00 UNLIMITED1 >>> (Priority) >>> account 181386lowmem testbatc user2 PENDING 0:00 UNLIMITED1 >>> (Priority) >>> account 181387lowmem testbatc user2 PENDING 0:00 UNLIMITED1 >>> (Priority) >>> >>> >>> We have tried to create new users and new accounts this afternoon and all >>> of them show (null) as their account when we break out the formatting rules >>> on sacct. >>> >>> sacctmgr add account accountname >>> sacctmgr add user username defaultaccount accountname >>> >>> We have even one case where all users under and account are working fine >>> except a user we added yesterday... so at some point in the past (logs >>> aren't helping us thus far) the ability to actually sync up a user and an >>> account for accounting purposes has left us. Also I have failed to mention >>> to this point that we are still running Slurm 2.5.4, my apologies for that. >>> >>> AC >>> >>> >>> On 08/23/2013 11:22 AM, Alan V. Cowles wrote: Sorry to spam the list, but we wanted to keep updates in flux. We managed to find the issue in our mysqldb we are using for job accounting which had the column value set to smallint (5) for that value, so it was rounding things off, some SQL magic and we now have appropriate uid's showing up. A new monkey wrench, some test jobs submitted by user3 below get their fairshare value of 5000 as expected, just not user2... we just cleared his jobs from the queue, and submitted another 100 jobs for testing and none of them got a fairshare value... In his entire history of using our cluster he hasn't submitted over 5000 jobs, in fact: [root@slurm-master ~]# sacct -c --format=user,jobid,jobname,start,elapsed,state,exitcode -u user2 | grep user2 | wc -l 2573 So we can't figure out why he's being overlooked. AC On 08/23/2013 10:31 AM, Alan V. Cowles wrote: > We think we may be onto something, in sacct we were looking at the jobs > submitted by the users, and found that many users share the same > uidnumber in the slurm database. It seems to correlate with the size of > the user's uid number in our ldap directory... users who's uid number are > greater than 65535 get trunked to that number... users with uid numbers > below that keep their correct uidnumbers (user2 in the sample output > below) > > > > > [root@slurm-master ~]# sacct -c > --format=User
[slurm-dev] Re: fairshare incrementing
Hey guys, We're still hung on our priority scheduling here, but I had a thought reading some other mailings to the list this morning. The syntax that others are using with user creation is not as basic as ours, and has other variables in place suchas "fairshare=parent" are these things that we need to specify when creating an account or are they defaults, wondering if this is why our newer users aren't showing up with the correct accounts in squeue. It still bugs us that accounts show up correctly in sacctmgr, just not in squeue for the point of enforcing priority. Could this be a bug corrected in later releases? AC On 08/23/2013 03:13 PM, Alan V. Cowles wrote: Final update for the day, we have found what is causing priority to be overlooked we just don't know what is causing it... [root@cluster-login ~]# squeue --format="%a %.7i %.9P %.8j %.8u %.8T %.10M %.9l %.6D %R" |grep user1 (null) 181378lowmem testbatc user1 PENDING 0:00 UNLIMITED1 (Priority) (null) 181379lowmem testbatc user1 PENDING 0:00 UNLIMITED1 (Priority) (null) 181380lowmem testbatc user1 PENDING 0:00 UNLIMITED1 (Priority) (null) 181381lowmem testbatc user1 PENDING 0:00 UNLIMITED1 (Priority) (null) 181382lowmem testbatc user1 PENDING 0:00 UNLIMITED1 (Priority) (null) 181383lowmem testbatc user1 PENDING 0:00 UNLIMITED1 (Priority) (null) 181384lowmem testbatc user1 PENDING 0:00 UNLIMITED1 (Priority) (null) 181385lowmem testbatc user1 PENDING 0:00 UNLIMITED1 (Priority) (null) 181386lowmem testbatc user1 PENDING 0:00 UNLIMITED1 (Priority) (null) 181387lowmem testbatc user1 PENDING 0:00 UNLIMITED1 (Priority) Compared to: [root@cluster-login ~]# squeue --format="%a %.7i %.9P %.8j %.8u %.8T %.10M %.9l %.6D %R" |grep user2 account 181378lowmem testbatc user2 PENDING 0:00 UNLIMITED 1 (Priority) account 181379lowmem testbatc user2 PENDING 0:00 UNLIMITED 1 (Priority) account 181380lowmem testbatc user2 PENDING 0:00 UNLIMITED 1 (Priority) account 181381lowmem testbatc user2 PENDING 0:00 UNLIMITED 1 (Priority) account 181382lowmem testbatc user2 PENDING 0:00 UNLIMITED 1 (Priority) account 181383lowmem testbatc user2 PENDING 0:00 UNLIMITED 1 (Priority) account 181384lowmem testbatc user2 PENDING 0:00 UNLIMITED 1 (Priority) account 181385lowmem testbatc user2 PENDING 0:00 UNLIMITED 1 (Priority) account 181386lowmem testbatc user2 PENDING 0:00 UNLIMITED 1 (Priority) account 181387lowmem testbatc user2 PENDING 0:00 UNLIMITED 1 (Priority) We have tried to create new users and new accounts this afternoon and all of them show (null) as their account when we break out the formatting rules on sacct. sacctmgr add account accountname sacctmgr add user username defaultaccount accountname We have even one case where all users under and account are working fine except a user we added yesterday... so at some point in the past (logs aren't helping us thus far) the ability to actually sync up a user and an account for accounting purposes has left us. Also I have failed to mention to this point that we are still running Slurm 2.5.4, my apologies for that. AC On 08/23/2013 11:22 AM, Alan V. Cowles wrote: Sorry to spam the list, but we wanted to keep updates in flux. We managed to find the issue in our mysqldb we are using for job accounting which had the column value set to smallint (5) for that value, so it was rounding things off, some SQL magic and we now have appropriate uid's showing up. A new monkey wrench, some test jobs submitted by user3 below get their fairshare value of 5000 as expected, just not user2... we just cleared his jobs from the queue, and submitted another 100 jobs for testing and none of them got a fairshare value... In his entire history of using our cluster he hasn't submitted over 5000 jobs, in fact: [root@slurm-master ~]# sacct -c --format=user,jobid,jobname,start,elapsed,state,exitcode -u user2 | grep user2 | wc -l 2573 So we can't figure out why he's being overlooked. AC On 08/23/2013 10:31 AM, Alan V. Cowles wrote: We think we may be onto something, in sacct we were looking at the jobs submitted by the users, and found that many users share the same uidnumber in the slurm database. It seems to correlate with the size of the user's uid number in our ldap directory... users who's uid number are greater than 65535 get trunked to that number... users with uid numbers below that keep their correct uidnumbers (user2 in the sample output below) [root@slurm-master ~]# sacct -c --format=User,uid,JobID,JobName,NodeList,Start,Elapsed,ExitCode,DerivedExitCode,state |grep user2|head user2 27545 30548 bwa node01-1 2013-07-08T13:04:25 00:00:4
[slurm-dev] Re: fairshare incrementing
Hey guys, We had a eureka moment, and have discovered the cause of the problem and how to get it back working. Now we need to prevent it from occurring again in the future. Looking through the mysql lead us to dead ends, as well as every sacctmgr thing we ran for the first couple of days. Finally we dug around in the transactions log, saw that there was a restart of slurmctld immediately after the last correctly functioning account was created. We actually wondered if you had to bounce the daemon in order to make priority work for users. Trying to find more info that we could cull from the command line I also attempted to run sview on our master node, and found it couldn't launch... it was then I discovered that if I ran scontrol ping, we got the following response: Slurmctld(primary/backup) at slurm-master/slurm-backup are DOWN/UP * ** RESTORE SLURMCTLD DAEMON TO SERVICE ** * Curious we decided to restart it on the master node, and we immediately entered a panic where it took a bit to re-import all of the jobs we have currently in our queue, but after a few minutes, we are currently in this status: Slurmctld(primary/backup) at slurm-master/slurm-backup are UP/DOWN * ** RESTORE SLURMCTLD DAEMON TO SERVICE ** * So from what we can tell, on or about june 17, the slurmctld on our master host failed and the backup host took off running with it. Meanwhile slurmdbd continued to run unfettered on our master node. For what we can figure, the slurmctld on the backup node could not communicate properly with the slurmdbd on the master as it was a remote host. Once we restored slurmctld to running on the master node with slurmdbd on the same host, priority fairshare began to work for all of our problematic users. Now we want to find a way to make sure this doesn't happen again to perhaps allow slurmdbd to also run on the backup node, or have the backup node be able to make a remote call to the dbd host? In slurmdbd.conf on the master node we have the following: # slurmDBD info DbdAddr=localhost DbdHost=localhost Would it help to put the ipaddr of the host itself and start slurmdbd on the backup node as well? On this page for ubuntu: http://manpages.ubuntu.com/manpages/jaunty/man5/slurmdbd.conf.5.html We found a reference to a slurmdbdaddr value that should be placed in slurm.conf and possible tell the slurmctld where the slurmdbd is running. Though most references to it seem to be ancient (slurmdbd 1.3?) and perhaps this is no longer needed in slurm 2.5.4+ Any thoughts on config modifications we could make? Thanks in advance. AC On 08/27/2013 11:24 AM, Alan V. Cowles wrote: Hey guys, We're still hung on our priority scheduling here, but I had a thought reading some other mailings to the list this morning. The syntax that others are using with user creation is not as basic as ours, and has other variables in place suchas "fairshare=parent" are these things that we need to specify when creating an account or are they defaults, wondering if this is why our newer users aren't showing up with the correct accounts in squeue. It still bugs us that accounts show up correctly in sacctmgr, just not in squeue for the point of enforcing priority. Could this be a bug corrected in later releases? AC On 08/23/2013 03:13 PM, Alan V. Cowles wrote: Final update for the day, we have found what is causing priority to be overlooked we just don't know what is causing it... [root@cluster-login ~]# squeue --format="%a %.7i %.9P %.8j %.8u %.8T %.10M %.9l %.6D %R" |grep user1 (null) 181378lowmem testbatc user1 PENDING 0:00 UNLIMITED1 (Priority) (null) 181379lowmem testbatc user1 PENDING 0:00 UNLIMITED1 (Priority) (null) 181380lowmem testbatc user1 PENDING 0:00 UNLIMITED1 (Priority) (null) 181381lowmem testbatc user1 PENDING 0:00 UNLIMITED1 (Priority) (null) 181382lowmem testbatc user1 PENDING 0:00 UNLIMITED1 (Priority) (null) 181383lowmem testbatc user1 PENDING 0:00 UNLIMITED1 (Priority) (null) 181384lowmem testbatc user1 PENDING 0:00 UNLIMITED1 (Priority) (null) 181385lowmem testbatc user1 PENDING 0:00 UNLIMITED1 (Priority) (null) 181386lowmem testbatc user1 PENDING 0:00 UNLIMITED1 (Priority) (null) 181387lowmem testbatc user1 PENDING 0:00 UNLIMITED1 (Priority) Compared to: [root@cluster-login ~]# squeue --format="%a %.7i %.9P %.8j %.8u %.8T %.10M %.9l %.6D %R" |grep user2 account 181378lowmem testbatc user2 PENDING 0:00 UNLIMITED1 (Priority) account 181379lowmem testbatc user2 PENDING 0:00 UNLIMITED1 (Priority) account 181380lowmem testbatc user2 PENDING 0:00 UNLIMITED1 (Priority) acco
[slurm-dev] RE: fairshare allocations
> -Original Message- > From: Bill Wichser [mailto:b...@princeton.edu] > Sent: Wednesday, January 21, 2015 5:20 AM > To: slurm-dev > Subject: [slurm-dev] fairshare allocations > > > The algorithm I use is fairtree under 14.11 but I believe that my > question relates to any method. > > As a University, we have many investments into a given cluster. At the > most simplistic level, lets assume there are but two two allocations. > The method I have been using is to assign a value, as a percentage of > ownership, to the various ACCOUNTs such that when summed across all > accounts, they add to 100. > > So chemistry might have a fairshare value of 20 as they contributed 20% > of the funding. Physics has a value of 10. And so forth, with many > having a fairshare value of 1 since no money was contributed. > > In the past, I simply assigned either a fairshare value of parent to the > users or assigned them a value of 1. > > So lets take a user, call him Bill, who has a fairshare value of 1 under > the account=chem. It appears to me that this 1 share is actually a 1 > share of the total and not a 1 share of what the account=chem owns. Am > I reading this correctly here? A share of 1 for Bill is a share of the total shares assigned to users (or accounts) under the chem account. Chem can have 1000 users, each with 1 share, but chem users' combined usage of the system will be throttled to 20% based on job priorities calculated by the fair-share factor. That works both ways: if only one user from chem is submitting jobs, that user can receive 20% of the resources of the cluster, even though they have only one share of chem. The most common practice is to assign a share of 1 to every user in an account. You can assign greater share values to users who are entitled to more than their peers. Don Lipari > > Thanks, > Bill
[slurm-dev] RE: fairshare allocations
On 01/21/2015 11:07 AM, Lipari, Don wrote: -Original Message- From: Bill Wichser [mailto:b...@princeton.edu] Sent: Wednesday, January 21, 2015 5:20 AM To: slurm-dev Subject: [slurm-dev] fairshare allocations The algorithm I use is fairtree under 14.11 but I believe that my question relates to any method. As a University, we have many investments into a given cluster. At the most simplistic level, lets assume there are but two two allocations. The method I have been using is to assign a value, as a percentage of ownership, to the various ACCOUNTs such that when summed across all accounts, they add to 100. So chemistry might have a fairshare value of 20 as they contributed 20% of the funding. Physics has a value of 10. And so forth, with many having a fairshare value of 1 since no money was contributed. In the past, I simply assigned either a fairshare value of parent to the users or assigned them a value of 1. So lets take a user, call him Bill, who has a fairshare value of 1 under the account=chem. It appears to me that this 1 share is actually a 1 share of the total and not a 1 share of what the account=chem owns. Am I reading this correctly here? A share of 1 for Bill is a share of the total shares assigned to users (or accounts) under the chem account. Chem can have 1000 users, each with 1 share, but chem users' combined usage of the system will be throttled to 20% based on job priorities calculated by the fair-share factor. That works both ways: if only one user from chem is submitting jobs, that user can receive 20% of the resources of the cluster, even though they have only one share of chem. The most common practice is to assign a share of 1 to every user in an account. You can assign greater share values to users who are entitled to more than their peers. Don Lipari Thanks, Bill So that was my expectation. But lets look at this account, truncated, with a user with a fairshare of 20 (using sshare -a -l -A ee -p) Account|User|Raw Shares|Norm Shares|Raw Usage|Norm Usage|Effectv Usage|FairShare|Level FS|GrpCPUMins|CPURunMins| ee||261|0.218227|189272064|0.047197|0.047197||4.623757||50912| ee|user1|1|0.009091|24151307|0.006022|0.127601|0.771261|0.071245||24605| ee|user2|1|0.009091|652289|0.000163|0.003446|0.780059|2.637872||0| ee|user3|25|0.227273|15684228|0.003911|0.082866|0.781525|2.742652||0| ... So ee as an account gets fairshare=261 and gets a 0.218227 normalized share count. A user underneath gets the expected 0.009091 normalized shares since there are a lot of fairshare=1 users there. The user3 gets basically 25x this value as the fairshare for user3=25 Yet the normalized shares is actually MORE than the normalized shares for the account as a whole. What should I make of this? Bill
[slurm-dev] RE: fairshare allocations
> -Original Message- > From: Bill Wichser [mailto:b...@princeton.edu] > Sent: Wednesday, January 21, 2015 8:23 AM > To: slurm-dev > Subject: [slurm-dev] RE: fairshare allocations > > > > > On 01/21/2015 11:07 AM, Lipari, Don wrote: > > > > > >> -Original Message- > >> From: Bill Wichser [mailto:b...@princeton.edu] > >> Sent: Wednesday, January 21, 2015 5:20 AM > >> To: slurm-dev > >> Subject: [slurm-dev] fairshare allocations > >> > >> > >> The algorithm I use is fairtree under 14.11 but I believe that my > >> question relates to any method. > >> > >> As a University, we have many investments into a given cluster. At the > >> most simplistic level, lets assume there are but two two allocations. > >> The method I have been using is to assign a value, as a percentage of > >> ownership, to the various ACCOUNTs such that when summed across all > >> accounts, they add to 100. > >> > >> So chemistry might have a fairshare value of 20 as they contributed 20% > >> of the funding. Physics has a value of 10. And so forth, with many > >> having a fairshare value of 1 since no money was contributed. > >> > >> In the past, I simply assigned either a fairshare value of parent to > the > >> users or assigned them a value of 1. > >> > >> So lets take a user, call him Bill, who has a fairshare value of 1 > under > >> the account=chem. It appears to me that this 1 share is actually a 1 > >> share of the total and not a 1 share of what the account=chem owns. Am > >> I reading this correctly here? > > > > A share of 1 for Bill is a share of the total shares assigned to users > > (or accounts) under the chem account. Chem can have 1000 users, each > with > > 1 share, but chem users' combined usage of the system will be throttled > > to 20% based on job priorities calculated by the fair-share factor. > > > > That works both ways: if only one user from chem is submitting jobs, > that > > user can receive 20% of the resources of the cluster, even though they > have > > only one share of chem. > > > > The most common practice is to assign a share of 1 to every user in an > > account. You can assign greater share values to users who are entitled > > to more than their peers. > > > > Don Lipari > > > >> > >> Thanks, > >> Bill > > So that was my expectation. But lets look at this account, truncated, > with a user with a fairshare of 20 (using sshare -a -l -A ee -p) > > Account|User|Raw Shares|Norm Shares|Raw Usage|Norm Usage|Effectv > Usage|FairShare|Level FS|GrpCPUMins|CPURunMins| > ee||261|0.218227|189272064|0.047197|0.047197||4.623757||50912| > > ee|user1|1|0.009091|24151307|0.006022|0.127601|0.771261|0.071245||24605| > ee|user2|1|0.009091|652289|0.000163|0.003446|0.780059|2.637872||0| > ee|user3|25|0.227273|15684228|0.003911|0.082866|0.781525|2.742652||0| > ... > > > > > > > So ee as an account gets fairshare=261 and gets a 0.218227 normalized > share count. > > A user underneath gets the expected 0.009091 normalized shares since > there are a lot of fairshare=1 users there. The user3 gets basically > 25x this value as the fairshare for user3=25 > > Yet the normalized shares is actually MORE than the normalized shares > for the account as a whole. What should I make of this? That looks like a bug. I don't see that behavior on our systems running slurm 14.03.11. Don > Bill
[slurm-dev] RE: fairshare allocations
On 01/21/2015 09:23 AM, Bill Wichser wrote: A user underneath gets the expected 0.009091 normalized shares since there are a lot of fairshare=1 users there. The user3 gets basically 25x this value as the fairshare for user3=25 Yet the normalized shares is actually MORE than the normalized shares for the account as a whole. What should I make of this? This is actually by design in Fair Tree and is different from other algorithms. The manpage for sshare covers this under "FAIR_TREE MODIFICATIONS".The manpage states that Norm Shares is "The shares assigned to the user or account normalized to the total number of assigned shares within the level." Basically, the Norm Shares is the association's raw shares value divided by the sum of it and its sibling associations' assigned raw shares values. For example, if an account has 10 users, each having 1 assigned raw share, the Norm Shares value will be .1 for each of those users under Fair Tree. Fair Tree only uses Norm Shares and Effective Usage (the other sshare field that's modified) when comparing sibling associations. Our Slurm UG presentation slides also mention this on pages 35 and 76 (http://slurm.schedmd.com/SUG14/fair_tree.pdf). Ryan
[slurm-dev] RE: fairshare allocations
Okay, I get it now. All the shares under a given ACCOUNT add up to 1.0. The divvy, and why some are higher than the actual ACCOUNT's number, is merely the effect from this allocation amongst the users underneath. So if the ACCOUNT gets 20% (0.20 or nearly so), then all the users underneath, when summed, have a value of 1.0. Because a user here gets a bigger cut by assignment of a fairshare value >1 then just because that user has a value exceeding the ACCOUNT value is not to be confused with a fairshare exceeding that of it's parent. Only that the value of the parent's fairshare, that user gets this percentage of the cut. Got it! Thanks much, Bill On 1/21/2015 12:57 PM, Ryan Cox wrote: On 01/21/2015 09:23 AM, Bill Wichser wrote: A user underneath gets the expected 0.009091 normalized shares since there are a lot of fairshare=1 users there. The user3 gets basically 25x this value as the fairshare for user3=25 Yet the normalized shares is actually MORE than the normalized shares for the account as a whole. What should I make of this? This is actually by design in Fair Tree and is different from other algorithms. The manpage for sshare covers this under "FAIR_TREE MODIFICATIONS".The manpage states that Norm Shares is "The shares assigned to the user or account normalized to the total number of assigned shares within the level." Basically, the Norm Shares is the association's raw shares value divided by the sum of it and its sibling associations' assigned raw shares values. For example, if an account has 10 users, each having 1 assigned raw share, the Norm Shares value will be .1 for each of those users under Fair Tree. Fair Tree only uses Norm Shares and Effective Usage (the other sshare field that's modified) when comparing sibling associations. Our Slurm UG presentation slides also mention this on pages 35 and 76 (http://slurm.schedmd.com/SUG14/fair_tree.pdf). Ryan
[slurm-dev] Re: Fairshare points return rate
The way I understood it was that the full share of points would return in 7 days. Its a _HalfLife_ Quoting Christopher B Coffey : Hello, I think I have either a configuration problem, or I am not understanding something correctly. I have the following set for fairshare in slurm.conf: PriorityType=priority/multifactor PriorityDecayHalfLife=7-0 Yet, a users fairshare points, shown by sshare, do not return at the rate that I had envisioned. The points are returning extremely slowly (many weeks). The way I understood it was that the full share of points would return in 7 days. In my example below, a user would have full fairshare points when the number was 0.25. Account User Raw Shares Norm Shares Raw Usage Effectv Usage FairShare -- -- --- --- - -- root 1.00 116705341 1.00 0.50 root root 10.50 0 0.00 1.00 normal 10.50 116705341 1.00 0.25 normal abc1 10.006329 0 0.012658 0.25 normal abc2 10.006329 0 0.012658 0.25 normal abc3 10.006329 0 0.012658 0.25 normal abc4 10.00632927021063 0.241260 0.00 normal abc5 10.006329 43576 0.013027 0.240107 normal abc6 10.006329 0 0.012658 0.25 normal abc7 10.006329 1045805 0.021506 0.094868 normal abc8 10.006329 3 0.012658 0.24 normal abc9 10.006329 8131420 0.081451 0.000134 Also, I’m not sure I have the shares correctly configured now that I think about it. In general fairshare priority has been working, but the points returning issue has vexed me, thanks guys! Chris
[slurm-dev] RE: fairshare - memory resource allocation
Bill, As I understand the dilemma you presented, you want to maximize the utilization of node resources when running with Slurm configured for SelectType=select/cons_res. To do this, you would like to nudge users into requesting only the amount of memory they will need for their jobs. The nudge would be in the form of decreased fair-share priority for users' jobs that request only one core but lots of memory. I don't know of a way for Slurm to do this as it exists. I can only offer alternatives that have their pros and cons. One alternative would be to add memory usage support to the multifactor priority plugin. This would be a substantial undertaking as it touches code not just in multifactor/priority_multifactor.c but also in structures that are defined in common/assoc_mgr.h as well as sshare itself. A second less invasive option would be to redefine the multifactor/priority_multifactor.c's raw_usage to make it a configurable blend of cpu and memory usage. These changes could be more localized to the multifactor/priority_multifactor.c module. However you would have a harder time justifying a user's sshare report because the usage numbers would no longer track jobs' historical cpu usage. You response to a user who asked you to justify their sshare usage report would be, "trust me, it's right". A third alternative (as I'm sure you know) is to give up on perfectly packed nodes and make every 4G of memory requested cost 1 cpu of allocation. Perhaps there are other options, but those are the ones that immediately come to mind. Don Lipari > -Original Message- > From: Bill Wichser [mailto:b...@princeton.edu] > Sent: Friday, July 25, 2014 6:14 AM > To: slurm-dev > Subject: [slurm-dev] fairshare - memory resource allocation > > > I'd like to revisit this... > > > After struggling with memory allocations in some flavor of PBS for over > 20 years, it was certainly a wonderful thing to have cgroup support > right out of the box with Slurm. No longer do we have a shared node's > jobs eating all the memory and killing everything running there. But we > have found that there is a cost to this and that is a failure to > adequately feed back this information to the fairshare mechanism. > > In looking at running jobs over the past 4 months, we found a spot where > we could reduce the DefMemPerCPU allocation in slurm.conf to a value > about 1G less than the actual G/core available. This meant that we had > to notify the users close to this max value so that they could adjust > their scripts. We also notified users that if this value was too high > that they'd do best to reduce that limit to exactly what they require. > This has proven much less successful. > > So our default is 3G/core with an actual node having 4G/core available. > This allows some bigger memory jobs and some smaller memory jobs to > make use of the node as there are available cores but not enough memory > for the default case. > > Now that is good. It allows higher utilization of nodes, all the while > protecting the memory of each other's processes. But the problem of > fairshare comes about pretty quickly when there are jobs requiring say > half the node's memory. This is mostly serial jobs requesting a single > core. So this leaves about 11 cores with only about 2G/core left. > Worse, when it comes to fairshare calculations it appears that these > jobs are only using a single core when in fact they are using half a > node. You can see where this is causing issues. > > Fairshare has a number of other issues as well, which I will send under > a different email. > > Now maybe this is just a matter of constant monitoring of user jobs and > proactively going after those users having small memory per core > requirements. We have attempted this in the past and have found that > the first job they run which crashes due to insufficient memory results > in all scripts being increased and so the process is never ending. > > Another solution is to simply trust the users and just keep reminding > them about allocations. They are usually a smart bunch and are quite > creative when it comes to getting jobs to run! So maybe I am concerned > over nothing at all and things will just work out. > > Bill
[slurm-dev] RE: fairshare - memory resource allocation
Bill and Don, We have wondered about this ourselves. I just came up with this idea and haven't thought it through completely, but option two seems like the easiest. For example, you could modify lines like https://github.com/SchedMD/slurm/blob/8a1e1384bacf690aed4c1f384da77a0cd978a63f/src/plugins/priority/multifactor/priority_multifactor.c#L952 to have a MAX() of a few different types. I seem to recall seeing this on the list or in a bug report somewhere already, but you could have different charge rates for memory or GPUs compared to a CPU, maybe on a per partition basis. You could give each of them a charge rate like: PartitionName=p1 ChargePerCPU=1.0 ChargePerGB=0.5 ChargePerGPU=2.0 .. So the line I referenced would be something like the following (except using real code and real struct members, etc): real_decay = run_decay * MAX(CPUs*ChargePerCPU, TotalJobMemory*ChargePerGB, GPUs*ChargePerGPU); In this case, each CPU is 1.0 but each GB of RAM is 0.5. Assuming no GPUs used, if the user requests 1 CPU and 2 GB of RAM the resulting usage is 1.0. But if they use 4 GB of RAM and 1 CPU, it is 2.0 just like they had been using 2 CPUs. Essentially you define every 2 GB of RAM to be equal to 1 CPU, so raw_usage could be redefined to deal with "cpu equivalents". It might be harder to explain to users but I don't think it would be too bad. Ryan On 07/25/2014 10:05 AM, Lipari, Don wrote: Bill, As I understand the dilemma you presented, you want to maximize the utilization of node resources when running with Slurm configured for SelectType=select/cons_res. To do this, you would like to nudge users into requesting only the amount of memory they will need for their jobs. The nudge would be in the form of decreased fair-share priority for users' jobs that request only one core but lots of memory. I don't know of a way for Slurm to do this as it exists. I can only offer alternatives that have their pros and cons. One alternative would be to add memory usage support to the multifactor priority plugin. This would be a substantial undertaking as it touches code not just in multifactor/priority_multifactor.c but also in structures that are defined in common/assoc_mgr.h as well as sshare itself. A second less invasive option would be to redefine the multifactor/priority_multifactor.c's raw_usage to make it a configurable blend of cpu and memory usage. These changes could be more localized to the multifactor/priority_multifactor.c module. However you would have a harder time justifying a user's sshare report because the usage numbers would no longer track jobs' historical cpu usage. You response to a user who asked you to justify their sshare usage report would be, "trust me, it's right". A third alternative (as I'm sure you know) is to give up on perfectly packed nodes and make every 4G of memory requested cost 1 cpu of allocation. Perhaps there are other options, but those are the ones that immediately come to mind. Don Lipari -Original Message- From: Bill Wichser [mailto:b...@princeton.edu] Sent: Friday, July 25, 2014 6:14 AM To: slurm-dev Subject: [slurm-dev] fairshare - memory resource allocation I'd like to revisit this... After struggling with memory allocations in some flavor of PBS for over 20 years, it was certainly a wonderful thing to have cgroup support right out of the box with Slurm. No longer do we have a shared node's jobs eating all the memory and killing everything running there. But we have found that there is a cost to this and that is a failure to adequately feed back this information to the fairshare mechanism. In looking at running jobs over the past 4 months, we found a spot where we could reduce the DefMemPerCPU allocation in slurm.conf to a value about 1G less than the actual G/core available. This meant that we had to notify the users close to this max value so that they could adjust their scripts. We also notified users that if this value was too high that they'd do best to reduce that limit to exactly what they require. This has proven much less successful. So our default is 3G/core with an actual node having 4G/core available. This allows some bigger memory jobs and some smaller memory jobs to make use of the node as there are available cores but not enough memory for the default case. Now that is good. It allows higher utilization of nodes, all the while protecting the memory of each other's processes. But the problem of fairshare comes about pretty quickly when there are jobs requiring say half the node's memory. This is mostly serial jobs requesting a single core. So this leaves about 11 cores with only about 2G/core left. Worse, when it comes to fairshare calculations it appears that these jobs are only using a single core when in fact they are using half a node. You can see where this is causing issues. Fairshare has a number of other issues as well, which I will send under a different email.
[slurm-dev] RE: fairshare - memory resource allocation
Thank you Ryan. Not sure how we will proceed here. Bill On 7/25/2014 12:30 PM, Ryan Cox wrote: Bill and Don, We have wondered about this ourselves. I just came up with this idea and haven't thought it through completely, but option two seems like the easiest. For example, you could modify lines like https://github.com/SchedMD/slurm/blob/8a1e1384bacf690aed4c1f384da77a0cd978a63f/src/plugins/priority/multifactor/priority_multifactor.c#L952 to have a MAX() of a few different types. I seem to recall seeing this on the list or in a bug report somewhere already, but you could have different charge rates for memory or GPUs compared to a CPU, maybe on a per partition basis. You could give each of them a charge rate like: PartitionName=p1 ChargePerCPU=1.0 ChargePerGB=0.5 ChargePerGPU=2.0 .. So the line I referenced would be something like the following (except using real code and real struct members, etc): real_decay = run_decay * MAX(CPUs*ChargePerCPU, TotalJobMemory*ChargePerGB, GPUs*ChargePerGPU); In this case, each CPU is 1.0 but each GB of RAM is 0.5. Assuming no GPUs used, if the user requests 1 CPU and 2 GB of RAM the resulting usage is 1.0. But if they use 4 GB of RAM and 1 CPU, it is 2.0 just like they had been using 2 CPUs. Essentially you define every 2 GB of RAM to be equal to 1 CPU, so raw_usage could be redefined to deal with "cpu equivalents". It might be harder to explain to users but I don't think it would be too bad. Ryan On 07/25/2014 10:05 AM, Lipari, Don wrote: Bill, As I understand the dilemma you presented, you want to maximize the utilization of node resources when running with Slurm configured for SelectType=select/cons_res. To do this, you would like to nudge users into requesting only the amount of memory they will need for their jobs. The nudge would be in the form of decreased fair-share priority for users' jobs that request only one core but lots of memory. I don't know of a way for Slurm to do this as it exists. I can only offer alternatives that have their pros and cons. One alternative would be to add memory usage support to the multifactor priority plugin. This would be a substantial undertaking as it touches code not just in multifactor/priority_multifactor.c but also in structures that are defined in common/assoc_mgr.h as well as sshare itself. A second less invasive option would be to redefine the multifactor/priority_multifactor.c's raw_usage to make it a configurable blend of cpu and memory usage. These changes could be more localized to the multifactor/priority_multifactor.c module. However you would have a harder time justifying a user's sshare report because the usage numbers would no longer track jobs' historical cpu usage. You response to a user who asked you to justify their sshare usage report would be, "trust me, it's right". A third alternative (as I'm sure you know) is to give up on perfectly packed nodes and make every 4G of memory requested cost 1 cpu of allocation. Perhaps there are other options, but those are the ones that immediately come to mind. Don Lipari -Original Message- From: Bill Wichser [mailto:b...@princeton.edu] Sent: Friday, July 25, 2014 6:14 AM To: slurm-dev Subject: [slurm-dev] fairshare - memory resource allocation I'd like to revisit this... After struggling with memory allocations in some flavor of PBS for over 20 years, it was certainly a wonderful thing to have cgroup support right out of the box with Slurm. No longer do we have a shared node's jobs eating all the memory and killing everything running there. But we have found that there is a cost to this and that is a failure to adequately feed back this information to the fairshare mechanism. In looking at running jobs over the past 4 months, we found a spot where we could reduce the DefMemPerCPU allocation in slurm.conf to a value about 1G less than the actual G/core available. This meant that we had to notify the users close to this max value so that they could adjust their scripts. We also notified users that if this value was too high that they'd do best to reduce that limit to exactly what they require. This has proven much less successful. So our default is 3G/core with an actual node having 4G/core available. This allows some bigger memory jobs and some smaller memory jobs to make use of the node as there are available cores but not enough memory for the default case. Now that is good. It allows higher utilization of nodes, all the while protecting the memory of each other's processes. But the problem of fairshare comes about pretty quickly when there are jobs requiring say half the node's memory. This is mostly serial jobs requesting a single core. So this leaves about 11 cores with only about 2G/core left. Worse, when it comes to fairshare calculations it appears that these jobs are only using a single core when in fact they are using half a node. You can see where th
[slurm-dev] RE: fairshare - memory resource allocation
Hi, As a variation on the second option you propose, take a look at the concept of Dominant Resource Fairness [1], which is an algorithm for achieving multi-resource (e.g. cpu's, memory, disk/net BW, ...) fairness. By using "dominant share"-secs instead of cpu-secs in the current accounting code the changes would similarly be limited in scope. [1] http://static.usenix.org/event/nsdi11/tech/full_papers/Ghodsi.pdf https://www.usenix.org/legacy/events/nsdi11/tech/slides/ghodsi.pdf -- Janne Blomqvist From: Lipari, Don [lipa...@llnl.gov] Sent: Friday, July 25, 2014 19:04 To: slurm-dev Subject: [slurm-dev] RE: fairshare - memory resource allocation Bill, As I understand the dilemma you presented, you want to maximize the utilization of node resources when running with Slurm configured for SelectType=select/cons_res. To do this, you would like to nudge users into requesting only the amount of memory they will need for their jobs. The nudge would be in the form of decreased fair-share priority for users' jobs that request only one core but lots of memory. I don't know of a way for Slurm to do this as it exists. I can only offer alternatives that have their pros and cons. One alternative would be to add memory usage support to the multifactor priority plugin. This would be a substantial undertaking as it touches code not just in multifactor/priority_multifactor.c but also in structures that are defined in common/assoc_mgr.h as well as sshare itself. A second less invasive option would be to redefine the multifactor/priority_multifactor.c's raw_usage to make it a configurable blend of cpu and memory usage. These changes could be more localized to the multifactor/priority_multifactor.c module. However you would have a harder time justifying a user's sshare report because the usage numbers would no longer track jobs' historical cpu usage. You response to a user who asked you to justify their sshare usage report would be, "trust me, it's right". A third alternative (as I'm sure you know) is to give up on perfectly packed nodes and make every 4G of memory requested cost 1 cpu of allocation. Perhaps there are other options, but those are the ones that immediately come to mind. Don Lipari > -Original Message- > From: Bill Wichser [mailto:b...@princeton.edu] > Sent: Friday, July 25, 2014 6:14 AM > To: slurm-dev > Subject: [slurm-dev] fairshare - memory resource allocation > > > I'd like to revisit this... > > > After struggling with memory allocations in some flavor of PBS for over > 20 years, it was certainly a wonderful thing to have cgroup support > right out of the box with Slurm. No longer do we have a shared node's > jobs eating all the memory and killing everything running there. But we > have found that there is a cost to this and that is a failure to > adequately feed back this information to the fairshare mechanism. > > In looking at running jobs over the past 4 months, we found a spot where > we could reduce the DefMemPerCPU allocation in slurm.conf to a value > about 1G less than the actual G/core available. This meant that we had > to notify the users close to this max value so that they could adjust > their scripts. We also notified users that if this value was too high > that they'd do best to reduce that limit to exactly what they require. > This has proven much less successful. > > So our default is 3G/core with an actual node having 4G/core available. > This allows some bigger memory jobs and some smaller memory jobs to > make use of the node as there are available cores but not enough memory > for the default case. > > Now that is good. It allows higher utilization of nodes, all the while > protecting the memory of each other's processes. But the problem of > fairshare comes about pretty quickly when there are jobs requiring say > half the node's memory. This is mostly serial jobs requesting a single > core. So this leaves about 11 cores with only about 2G/core left. > Worse, when it comes to fairshare calculations it appears that these > jobs are only using a single core when in fact they are using half a > node. You can see where this is causing issues. > > Fairshare has a number of other issues as well, which I will send under > a different email. > > Now maybe this is just a matter of constant monitoring of user jobs and > proactively going after those users having small memory per core > requirements. We have attempted this in the past and have found that > the first job they run which crashes due to insufficient memory results > in all scripts being increased and so the process is never ending. > > Another solution is to simply trust the users and just keep reminding > them about allocations. They are usually a smart bunch and are quite > creative when it comes to getting jobs to run! So maybe I am concerned > over nothing at all and things will just work out. > > Bill
[slurm-dev] RE: fairshare - memory resource allocation
I'm interested in hearing opinions on this, if any. Basically, I think there is an easy solution to the problem of a user using few CPUs but a lot of memory and that not being reflected well in the CPU-centric usage stats. Below is my proposal. There are likely some other good approaches out there too (Don and Janne presented some) so feel free to tell me that you don't like this idea :) Short version I propose that the Raw Usage be modified to *optionally* be ("CPU equivalents" * time) instead of just (CPUs * time). The "CPU equivalent" would be a MAX() of CPUs, memory, nodes, GPUs, energy over that time period, or whatever multiplied by a corresponding charge rate that an admin can configure on a per partition basis. I wrote a simple proof of concept patch to demonstrate this (see "Proof of Concept" below for details). Longer version The CPU equivalent would be used in place of total_cpus for calculating usage_raw. I propose that the default charge rate be 1.0 for each CPU in a job and 0.0 for everything else. This is the current behavior so there are no behavior changes if you choose not to define a different charge rate. The reason I think this should be done on a partition basis is because different partitions may have nodes with different memory/core ratios, etc. so one partition may have 2 GB/core and another may have 8 GB/core nodes and you may want to charge differently on each. If you define the charge rate for each CPU to be 1.0 and the charge rate per GB of memory to be 0.5, that is saying that 2 GB of memory will be equivalent to the charge rate for 1 CPU. 4 GB of memory would be equivalent to 2 CPUs (4 GB * 0.5/GB). Since it is a MAX() of all the available (resource * charge_rate) combinations, the largest value is chosen. If a user uses 1 CPU and 1 TB of RAM out of a 1 TB node, the user gets charged for using all the RAM. If a user uses 16 CPUs and 1 MB, the user gets charged for 16 CPUs. Downsides The problem that is not completely solved is if a user uses 1 CPU but 3/4 of the memory on a node. Then they only get billed for 3/4 of the node but might make it unusable for others who need a whole or half node. I'm not sure of a great way to solve that besides modifying the request in a job submit plugin or requiring exclusive node access. One other complication is for resources that include a counter rather than a static allocation value, such as network bandwidth or energy. This is a problem because the current approach is to immediately begin decaying the cputime (aka usage) as it accumulates. This means you would have to keep a delta value for each resource with a counter, meaning you track that 5 GB have been transmitted since the last decay thread iteration then only add that 5 GB. This could get messy when comparing MAX(total_cpus * charge_per_cpu, total_bw * charge_bw_per_gb) each iteration since the bandwidth may never reach a high enough value to matter between iterations but might when considered as an entire job. I don't think this proposal would be too bad for something like energy. You could define a charge rate per joule (or kilojoule or whatever) that would equal the node's minimum power divided by core count. Then you look at the delta of that time period. If they were allocated all cores and used minimum power, they get charged 1.0 * core count. If they were allocated all cores and used maximum power, they effectively get charged for the difference in the node's max energy and min energy times the energy charge rate. This calculation, as with others, would occur once per decay thread iteration. User Education The reason I like this approach is that it is incredibly simple to implement and I don't think it takes much effort to explain to users. It would be easy to add other resources you want to charge for (it would require a code addition, though it would be pretty simple if the data is available in the right structs). It doesn't require any RPC changes. sshare, etc only need manpage clarifications to say that the usage data is "CPU equivalents". No new fields are required. As for user education, you just need to explain the concept of "CPU equivalents", something that can be easily done in the documentation. The slurm.conf partition lines would be relatively easy to read too. If you don't need to change the behavior, no slurm.conf changes or explanations to users are required. Proof of Concept I did a really quick proof of concept (attached) based on the master branch. It is very simple to charge for most things as long as the data is there in the existing structs. One caveat for the test patch is that I didn't see a float handler in the config parser so I skipped over that for the test. Instead, each config parameter in slurm.conf should be set to (desired_value * 1000) for now. Proper float handling can be added if this is the route people want to take.
[slurm-dev] RE: fairshare - memory resource allocation
Hi, if I understand it correctly, this is actually very close to Dominant Resource Fairness (DRF) which I mentioned previously, with the difference that in DRF the charge rates are determined automatically from the available resources (in a partition) rather than being specified explicitly by the administrator. So for an example, say we have a partition with 100 cores and 400 GB memory. Now for a job requesting (10CPU's, 20 GB) the domination calculation proceeds as follows: 1) Calculate the "domination vector" by dividing each element in the request vector (here, CPU & MEM) with the available resources. That is (10/100, 20/400) = (0.1, 0.05). 2) The MAX element in the domination vector is chosen (it "dominates" the others, hence the name of the algorithm) as the one to use in fairshare calculations, accounting etc. In this case, the CPU element (0.1). Now for another job request, (1CPU, 20 GB) the domination vector is (0.01, 0.05) and the MAX element is the memory element (0.05), so in this case the memory part of the request dominates. In your patch you have used "cpu-sec equivalents" rather than "dominant share secs", but that's just a difference of a scaling factor. From a backwards compatibility and user education point of view cpu-sec equivalents seem like a better choice to me, actually. So while you patch is more flexible than DRF in that it allows arbitrary charge rates to be specified, I'm not sure it makes sense to specify rates different from the DRF ones? Or if one does specify different rates, it might end up breaking some of the fairness properties that are described in the DRF paper and opens up the algorithm for gaming? -- Janne Blomqvist From: Ryan Cox [ryan_...@byu.edu] Sent: Tuesday, July 29, 2014 18:47 To: slurm-dev Subject: [slurm-dev] RE: fairshare - memory resource allocation I'm interested in hearing opinions on this, if any. Basically, I think there is an easy solution to the problem of a user using few CPUs but a lot of memory and that not being reflected well in the CPU-centric usage stats. Below is my proposal. There are likely some other good approaches out there too (Don and Janne presented some) so feel free to tell me that you don't like this idea :) Short version I propose that the Raw Usage be modified to *optionally* be ("CPU equivalents" * time) instead of just (CPUs * time). The "CPU equivalent" would be a MAX() of CPUs, memory, nodes, GPUs, energy over that time period, or whatever multiplied by a corresponding charge rate that an admin can configure on a per partition basis. I wrote a simple proof of concept patch to demonstrate this (see "Proof of Concept" below for details). Longer version The CPU equivalent would be used in place of total_cpus for calculating usage_raw. I propose that the default charge rate be 1.0 for each CPU in a job and 0.0 for everything else. This is the current behavior so there are no behavior changes if you choose not to define a different charge rate. The reason I think this should be done on a partition basis is because different partitions may have nodes with different memory/core ratios, etc. so one partition may have 2 GB/core and another may have 8 GB/core nodes and you may want to charge differently on each. If you define the charge rate for each CPU to be 1.0 and the charge rate per GB of memory to be 0.5, that is saying that 2 GB of memory will be equivalent to the charge rate for 1 CPU. 4 GB of memory would be equivalent to 2 CPUs (4 GB * 0.5/GB). Since it is a MAX() of all the available (resource * charge_rate) combinations, the largest value is chosen. If a user uses 1 CPU and 1 TB of RAM out of a 1 TB node, the user gets charged for using all the RAM. If a user uses 16 CPUs and 1 MB, the user gets charged for 16 CPUs. Downsides The problem that is not completely solved is if a user uses 1 CPU but 3/4 of the memory on a node. Then they only get billed for 3/4 of the node but might make it unusable for others who need a whole or half node. I'm not sure of a great way to solve that besides modifying the request in a job submit plugin or requiring exclusive node access. One other complication is for resources that include a counter rather than a static allocation value, such as network bandwidth or energy. This is a problem because the current approach is to immediately begin decaying the cputime (aka usage) as it accumulates. This means you would have to keep a delta value for each resource with a counter, meaning you track that 5 GB have been transmitted since the last decay thread iteration then only add that 5 GB. This could get messy when comparing MAX(total_cpus * charge_per_cpu, total_bw * charge_bw_per_gb) each iteration since the bandwidth may never reach a high enough value to matter between iterations but might when c
[slurm-dev] RE: fairshare - memory resource allocation
Just a short note about terminology. I believe "processor equivalents" (PE) is a much used term for this. It is at least what Maui and Moab uses, if I recall correctly. The "resource*time" would then be PE seconds (or hours, or whatever). -- Regards, Bjørn-Helge Mevik, dr. scient, Department for Research Computing, University of Oslo
[slurm-dev] RE: fairshare - memory resource allocation
Janne, I appreciate the feedback. I agree that it makes the most sense to specify rates like DRF most of the time. However, there are some use cases that I'm aware of and others that are probably out there that would make a DRF imitation difficult or less desirable if it's the only option. We happen to have one partition that has mixed memory amounts per node, 32 GB and 64 GB. Besides the memory differences (long story), the nodes are homogeneous and each have 16 cores. I'm not sure I would like the DRF approach for this particular scenario. In this case we would like to set the charge rate to be .5/GB, or 1 core == 2 GB RAM. If someone needs 64 GB per node, they are contending for a more limited resource and we would be happy to double the charge rate for the 64 GB nodes. If they need all 64 GB, they would end up being charged for 32 CPU/processor equivalents instead of 16. With DRF that wouldn't be possible if I understand correctly. One other feature that could be interesting is to have a "baseline" standard for a CPU charge on a per-partition basis. Let's say that you have three partitions: old_hardware, new_hardware, and super_cooled_overclocked_awesomeness. You could set the per CPU charges to be 0.8, 1.0, and 20.0. That would reflect that a cpu-hour on one partition doesn't result in the same amount of computation as in another partition. You could accomplish the same thing automatically by using a QOS (and maybe some other parameter I'm not aware of) and maybe a job submit plugin but this would make it easier. I don't know that we would do this in our setup but it would be possible. It would be possible to add a config parameter that is something like Mem=DRF that would auto-configure it to match. The one question I have about that approach is what to do about partitions with non-homogeneous nodes. Does it make sense to sum the total cores and memory, etc or should it default to a charge rate that is the min() of the node configurations? Of course, partitions with mixed node types could be difficult to support no matter what method is used for picking charge rates. So yes, having a DRF-like auto-configuration could be nice and we might even use it for most of our partitions. I don't think I'll attempt it for the initial implementation but we'll see. Thanks, Ryan On 07/30/2014 03:31 PM, Blomqvist Janne wrote: Hi, if I understand it correctly, this is actually very close to Dominant Resource Fairness (DRF) which I mentioned previously, with the difference that in DRF the charge rates are determined automatically from the available resources (in a partition) rather than being specified explicitly by the administrator. So for an example, say we have a partition with 100 cores and 400 GB memory. Now for a job requesting (10CPU's, 20 GB) the domination calculation proceeds as follows: 1) Calculate the "domination vector" by dividing each element in the request vector (here, CPU & MEM) with the available resources. That is (10/100, 20/400) = (0.1, 0.05). 2) The MAX element in the domination vector is chosen (it "dominates" the others, hence the name of the algorithm) as the one to use in fairshare calculations, accounting etc. In this case, the CPU element (0.1). Now for another job request, (1CPU, 20 GB) the domination vector is (0.01, 0.05) and the MAX element is the memory element (0.05), so in this case the memory part of the request dominates. In your patch you have used "cpu-sec equivalents" rather than "dominant share secs", but that's just a difference of a scaling factor. From a backwards compatibility and user education point of view cpu-sec equivalents seem like a better choice to me, actually. So while you patch is more flexible than DRF in that it allows arbitrary charge rates to be specified, I'm not sure it makes sense to specify rates different from the DRF ones? Or if one does specify different rates, it might end up breaking some of the fairness properties that are described in the DRF paper and opens up the algorithm for gaming? -- Janne Blomqvist ________________ From: Ryan Cox [ryan_...@byu.edu] Sent: Tuesday, July 29, 2014 18:47 To: slurm-dev Subject: [slurm-dev] RE: fairshare - memory resource allocation I'm interested in hearing opinions on this, if any. Basically, I think there is an easy solution to the problem of a user using few CPUs but a lot of memory and that not being reflected well in the CPU-centric usage stats. Below is my proposal. There are likely some other good approaches out there too (Don and Janne presented some) so feel free to tell me that you don't like this idea :) Short version I propose that the Raw Usage be modified to *optionally* be ("CPU equivalents" * time) instead of just (CPUs * time). Th
[slurm-dev] RE: fairshare - memory resource allocation
Thanks. I can certainly call it that. My understanding is that this would be a slightly different implementation from Moab/Maui, but I don't know those as well so I could be wrong. Either way, the concept is similar enough that a more recognizable term might be good. Does anyone else have thoughts on this? I called it "CPU equivalents" because the calculation in the code is currently ("total_cpus" * time) so I stuck with CPUs. Slurm seems to use lots of terms somewhat interchangeably so I couldn't really decide. I don't really have an opinion on the name so I'll just accept what others decide. Ryan On 07/31/2014 02:28 AM, Bjørn-Helge Mevik wrote: Just a short note about terminology. I believe "processor equivalents" (PE) is a much used term for this. It is at least what Maui and Moab uses, if I recall correctly. The "resource*time" would then be PE seconds (or hours, or whatever).
[slurm-dev] RE: fairshare - memory resource allocation
All, There has been more conversation on http://bugs.schedmd.com/show_bug.cgi?id=858. It might be good to post future comments there so we have just one central location for everything. No worries if you'd rather reply on the list. Once a solution is ready I'll post something to the list so everyone is aware. Ryan
[slurm-dev] RE: fairshare - memory resource allocation
Hi all, this is a very interesting approach. I hope we find a chance to discuss it in Lugano. Ulf -- ___ Dr. Ulf Markwardt Technische Universität Dresden Center for Information Services and High Performance Computing (ZIH) 01062 Dresden, Germany Phone: (+49) 351/463-33640 WWW: http://www.tu-dresden.de/zih smime.p7s Description: S/MIME Cryptographic Signature
[slurm-dev] Re: Fairshare factor not raising
Hello, I am still experiencing the same issue, which is described below. I would be glad to get any hint where to look for a solution. - Do you think it might help to recreate the database from scratch? Kind Regards and thank you all in advance! On 05/30/2017 02:02 PM, Simon Kreuzer wrote: Hello, After a crash of the machine on which slurmctld and slumdbd is running, fairshare factors as seen in /sshare/ are only decreasing for people submitting jobs. Even though the corresponding users didn't submit jobs for days, the fairshare factor stays zero. It seems like PriorityDecayHalfLife=7-0 does not have any effect. I restartet both, slurmctld and slurmdbd deamon and also tried resetting using /PriorityUsageResetPeriod=NOW. /It didn't change anything. Has anyone a hint where this issue might come from? It used to work before. The priority settings in slum.conf are the following: FastSchedule=1 SchedulerType=sched/backfill PriorityType=priority/multifactor PriorityDecayHalfLife=7-0 PriorityMaxAge=14-0 PriorityFavorSmall=YES PriorityFlags=SMALL_RELATIVE_TO_TIME PriorityWeightAge=5000 PriorityWeightFairshare=10 PriorityWeightJobSize=1000 PriorityWeightPartition=1 PriorityWeightQOS=0 PriorityUsageResetPeriod=NONE Thanks in advance!
[slurm-dev] RE: Fairshare=parent on an account: What should it do?
Hi Ryan, We currently use have a similar set up which may achieve what you want. We have parent accounts which are assigned a fairshare based upon the amount of funding they provide. In this set up we do not use fairshare=parent but rather that the individual user accounts have their group as the parent. Both the parent group and individual user accounts have a fairshare value. For example; group1 40 | ---> user_foo account 10 | | |---> user_foo 1 | |---> user_bar account 10 | ---> user_bar 1 group2 30 | ---> user account 10 | ---> user 1 group3 30 | ---> user account 10 | ---> user 1 The user accounts under group1 will have a higher priority from their fairshare factor than users from groups 2 or 3. As users are using their own individual child accounts there is also a fairshare in affect within the group, this allows for a user who has not run very much to be given higher priority than a larger user in the same group which prevents them from long queuing times. I have copied pasted some acct commands with a number of fields removed in the event the above ASCII diagram suffers a formatting error. sacctmgr show account group1 withasccoc Account Descr Par Name Share group1 group1 root 40 sacctmgr show account foo withasccoc Account Descr Par NameShare foo group1_foo group1 10 foo group1_foo foo 1 As far as account automation all our users exist within LDAP so we use a cron job to automatically poll LDAP and add users if they are not already present in slurm. You may be able to give the coordinator rights to the grad student but this may be more control than you want to give. Kind regards, George From: Ryan Cox [ryan_...@byu.edu] Sent: 11 June 2014 00:20 To: slurm-dev Subject: [slurm-dev] Fairshare=parent on an account: What should it do? We're trying to figure out what the intended behavior of Fairshare=parent is when set on an account (http://bugs.schedmd.com/show_bug.cgi?id=864). We know what the actual behavior is but we're wondering if anyone actually likes the current behavior. There could be some use case out there that we don't know about. For example, you can end up with a scenario like the following: acctProf / | \ / | \ acctTA(parent) uD(5) uE(5) / | \ / | \ uA(5) uB(5) uC(5) The number in parenthesis is Fairshare according to sacctmgr. We incorrectly thought that Fairshare=parent would essentially collapse the tree so that uA - uE would all be on the same level. Thus, all five users would each get 5 / 25 shares. What actually happens is you get the following shares at the user level: shares (uA) = 5 / 15 = .333 shares (uB) = 5 / 15 = .333 shares (uC) = 5 / 15 = .333 shares (uD) = 5 / 10 = .5 shares (uE) = 5 / 10 = .5 That's pretty far off from each other, but not as far as it would be if one account had two users and the other had forty. Assuming this demonstration value of 5 shares, that would be: user_in_small_account = 5 / (2*5) = .5 user_in_large_account = 5 / (40*5) = .025 Is that actually useful to someone? We want to use subaccounts below a faculty account to hold, for example, a grad student or postdoc who teaches a class. It would be nice for the grad student to have administrative control over the subaccount since he actually knows the students but not have it affect priority calculations. Ryan -- Ryan Cox Operations Director Fulton Supercomputing Lab Brigham Young University