Re: [lustre-discuss] Hidden QoS in Lustre ?
Dear Simon, Thank you very much for your useful information. Now we are arranging the system maintenance date in order to upgrade to Lustre-2.12.5. Then we will follow your suggestion to see whether this problem could be fixed. Here I report a test of under continuous I/O, how the cur_grant_bytes changed overtime. Again the client runs the following script for continuous reading in the background: # The Lustre file system was mounted under /home while [ 1 ]; do tar cf - /home/large/data | ssh remote_host "cat > /dev/null" done And every 20 mins, in the same client we copied a 600MB file from one directory to another within Lustre, and check the "cur_grant_bytes" by the following command running in the same client: /opt/lustre/sbin/lctl get_param osc.*.cur_grant_bytes The result is (every line separated by around 20 mins): osc.chome-OST-osc-88a03915.cur_grant_bytes=1880752127 osc.chome-OST-osc-88a03915.cur_grant_bytes=1410564096 osc.chome-OST-osc-88a03915.cur_grant_bytes=1059201024 osc.chome-OST-osc-88a03915.cur_grant_bytes=794400768 osc.chome-OST-osc-88a03915.cur_grant_bytes=595800576 osc.chome-OST-osc-88a03915.cur_grant_bytes=446850432 osc.chome-OST-osc-88a03915.cur_grant_bytes=335137824 osc.chome-OST-osc-88a03915.cur_grant_bytes=251353368 osc.chome-OST-osc-88a03915.cur_grant_bytes=188515026 osc.chome-OST-osc-88a03915.cur_grant_bytes=141386270 osc.chome-OST-osc-88a03915.cur_grant_bytes=106039703 osc.chome-OST-osc-88a03915.cur_grant_bytes=79529778 osc.chome-OST-osc-88a03915.cur_grant_bytes=59647334 osc.chome-OST-osc-88a03915.cur_grant_bytes=44735501 osc.chome-OST-osc-88a03915.cur_grant_bytes=33551626 osc.chome-OST-osc-88a03915.cur_grant_bytes=25163720 osc.chome-OST-osc-88a03915.cur_grant_bytes=18872790 osc.chome-OST-osc-88a03915.cur_grant_bytes=14154593 osc.chome-OST-osc-88a03915.cur_grant_bytes=10615945 osc.chome-OST-osc-88a03915.cur_grant_bytes=7961959 osc.chome-OST-osc-88a03915.cur_grant_bytes=5971470 osc.chome-OST-osc-88a03915.cur_grant_bytes=4478603 osc.chome-OST-osc-88a03915.cur_grant_bytes=3358953 osc.chome-OST-osc-88a03915.cur_grant_bytes=2519215 osc.chome-OST-osc-88a03915.cur_grant_bytes=1889412 osc.chome-OST-osc-88a03915.cur_grant_bytes=1417059 osc.chome-OST-osc-88a03915.cur_grant_bytes=1062795 osc.chome-OST-osc-88a03915.cur_grant_bytes=797097 osc.chome-OST-osc-88a03915.cur_grant_bytes=797097 The value 797097 seems to be the minimum. When it dropped to 1062795, the time of cp dramatically increased from around 1 sec to 1 min. In addition, during the test, the cluster is completely idling. And it is obvious that this test does not saturate the loading of both network and MDT / OST hardware (they have almost no loading). I am wondering whether this could be a bug to report to the development team. Best Regards, T.H.Hsieh On Thu, Oct 29, 2020 at 09:49:42AM -0400, Simon Guilbault wrote: > Our current workaround was to use the following command on the MGS with > Lustre 2.12.5 that include the patches in LU-12651 and LU-12759 (we were > using a patched 2.12.4 a few months ago): > lctl set_param -P osc.*.grant_shrink=0 > > We could not find the root cause of the underlying problem, dynamic grant > shrinking seems to be useful when the OSTs are running out of free space. > > On Wed, Oct 28, 2020 at 11:47 PM Tung-Han Hsieh < > thhs...@twcp1.phys.ntu.edu.tw> wrote: > > > Dear Simon, > > > > Thank you very much for your hint. Yes, you are right. We compared > > the grant size of two client by (running in each client): > > > > lctl get_param osc.*.cur_grant_bytes > > > > - Client A: It has run the following large data transfer for over 36 hrs. > > > > while [ 1 ]; do > > tar cf - /home/large/data | ssh remote_host "cat > /dev/null" > > done > > > > The value of "cur_grant_bytes" is 796134. > > > > - Client B: It is almost idling during the action of Client A. > > > > The value of "cur_grant_bytes" is 1715863552. > > > > If this is the reason that hit the I/O performance of Client A seriously, > > is it possible to maintain it at a constant value at least for the head > > node (since the head node is the most probable one to have large and long > > time data I/O of the whole cluster, especially for a data center) ? > > > > I would be also like to ask: Why this value has to be dynamically adjusted > > ? > > > > Thank you very much for your comment in advance. > > > > Best Regards, > > > > T.H.Hsieh > > > > On Wed, Oct 28, 2020 at 02:00:21PM -0400, Simon Guilbault wrote: > > > Hi, we had a similar performance problem on our login/DTNs node a few > > > months ago, the problem was the gran
Re: [lustre-discuss] Hidden QoS in Lustre ?
Our current workaround was to use the following command on the MGS with Lustre 2.12.5 that include the patches in LU-12651 and LU-12759 (we were using a patched 2.12.4 a few months ago): lctl set_param -P osc.*.grant_shrink=0 We could not find the root cause of the underlying problem, dynamic grant shrinking seems to be useful when the OSTs are running out of free space. On Wed, Oct 28, 2020 at 11:47 PM Tung-Han Hsieh < thhs...@twcp1.phys.ntu.edu.tw> wrote: > Dear Simon, > > Thank you very much for your hint. Yes, you are right. We compared > the grant size of two client by (running in each client): > > lctl get_param osc.*.cur_grant_bytes > > - Client A: It has run the following large data transfer for over 36 hrs. > > while [ 1 ]; do > tar cf - /home/large/data | ssh remote_host "cat > /dev/null" > done > > The value of "cur_grant_bytes" is 796134. > > - Client B: It is almost idling during the action of Client A. > > The value of "cur_grant_bytes" is 1715863552. > > If this is the reason that hit the I/O performance of Client A seriously, > is it possible to maintain it at a constant value at least for the head > node (since the head node is the most probable one to have large and long > time data I/O of the whole cluster, especially for a data center) ? > > I would be also like to ask: Why this value has to be dynamically adjusted > ? > > Thank you very much for your comment in advance. > > Best Regards, > > T.H.Hsieh > > On Wed, Oct 28, 2020 at 02:00:21PM -0400, Simon Guilbault wrote: > > Hi, we had a similar performance problem on our login/DTNs node a few > > months ago, the problem was the grant size was shrinking and was getting > > stuck under 1MB. Once under 1MB, the client had to send every request to > > the OST using sync IO. > > > > Check the output of the following command: > > lctl get_param osc.*.cur_grant_bytes > > > > On Wed, Oct 28, 2020 at 12:08 AM Tung-Han Hsieh < > > thhs...@twcp1.phys.ntu.edu.tw> wrote: > > > > > Dear All, > > > > > > Sorry that I am not sure whether this mail was successfully posted to > > > the lustre-discuss mailing list or not. So I resent it again. Please > > > ignore it if you already read it before. > > > > > > > === > > > > > > Dear Andreas, > > > > > > Thank you very much for your kindly suggestions. These days I got a > chance > > > to follow your suggestions for the test. This email is to report the > > > results > > > I have done so far. What I have done were: > > > > > > 1. Upgrade one client (with Infiniband) to Lustre 2.13.56_44_gf8a8d3f > > >(obtained from github). The compiling information is: > > > > > >- Linux kernel 4.19.123. > > >- Infiniband MLNX_OFED_SRC-4.6-1.0.1.1. > > >- ./configure --prefix=/opt/lustre \ > > > --with-o2ib=/path/of/mlnx-ofed-kernel-4.6 \ > > > --disable-server --enable-mpitests=no > > >- make > > >- make install > > > > > > 2. We mounted the lustre file system (lustre MDT/OST servers: version > > >2.12.4 with Infiniband with ZFS backend) by this command: > > > > > >- mount -t lustre -o flock mdt@o2ib:/chome /home > > > > > > 3. The script to simulate large data transfer is following: > > >(the directory "/home/large/data" contains 758 files, each size > 600MB) > > > > > >while [ 1 ]; do > > >tar cf - /home/large/data | ssh remote_host "cat > /dev/null" > > >done > > > > > >ps. Note that this scenario is common in a large data center, while > > >some users transferring large data out of the data center > through > > >the head node; while other users might copy files and do their > > >normal works in the same head node. > > > > > > 4. During the data transfer in the background, I occationally ran this > > >command in the same client to test whether there is any abnormality > > >in I/O performance (where /home/dir1/file has size 600MB): > > > > > >cp /home/dir1/file /home/dir2/ > > > > > >In the beginning this command can complete in about 1 sec. But after > > >around 18 hours (not exactly, because the test ran overnight while > > >I was sleeping), the problem appeared. The time to complete the same > > >cp command was more than 1 minute. > > > > > >During the test, I am sure that the whole cluster was idling. The > MDT > > >and OST servers did not have other loading. The CPU usage of the > testing > > >client was below 0.3. > > > > > >Then I stopped the test, and let the whole system completely idle. > But > > >after 3 hours, the I/O abnormality of the same "cp" command was > still > > >there. Only after I unmounted /home and remounted /home, the > abnormality > > >of "cp" recovered to normal. > > > > > > Before and after remounting /home (which I call "reset"), I did the > > > following tests: > > > > > > 1. Using "top" to check the memory usag
Re: [lustre-discuss] ZFS atime is it required?
On Oct 23, 2020, at 14:03, Kumar, Amit mailto:ahku...@mail.smu.edu>> wrote: Dear All, Quick question, can I get away by setting “zfs set atime=off on_all_my_voulmes_mgt_mdt_and_osts” ? I ask this as it is noted to be performance boosting tip with the assumption filesystems(Lustre) handles all access times? You don't really need atime enabled on the OSTs, but I also don't think "atime=off" will make any difference. That is a VFS/ZPL level option, and Lustre osd-zfs doesn't use any of the ZPL code, but rather handles atime internally. Cheers, Andreas -- Andreas Dilger Principal Lustre Architect Whamcloud ___ lustre-discuss mailing list lustre-discuss@lists.lustre.org http://lists.lustre.org/listinfo.cgi/lustre-discuss-lustre.org