Re: strange load values
On Thu, 12 Mar 2009, jmc wrote: --- Gabri Mate [Wed, Mar 11, 2009 at 08:58:15PM +0100]: --- Hey there! I have a Compaq DL580 G1 with 4x700Mhz PIII, 2GB RAM, 2x36GB U320 SCSI on a HP SmartArray 5300 with 64MB BBU. It's a general purpose 'hobby' server. The average concurrent i believe that Theo and many of the devs have said many times that the load average means nothing. here's a reference to one such thread: http://kerneltrap.org/mailarchive/openbsd-misc/2008/11/8/4041294 Perhaps the man page of w(1) should be changed. Currently it states: The load average numbers give the number of jobs in the run queue averaged over 1, 5 and 15 minutes. which is rather inaccurate or wrong. I suggest the patch below. Please correct my English if necessary. Regards, David Index: src/usr.bin/w/w.1 === RCS file: /cvs/src/usr.bin/w/w.1,v retrieving revision 1.18 diff -u -p -r1.18 w.1 --- src/usr.bin/w/w.1 31 May 2007 19:20:19 - 1.18 +++ src/usr.bin/w/w.1 13 Mar 2009 13:10:56 - @@ -50,7 +50,8 @@ The first line displays the current time been running, the number of users logged into the system, and the load averages. The load average numbers give the number of jobs in the run queue averaged -over 1, 5 and 15 minutes. +over past 1, 5 and 15 minutes. Because of the sampling method and +algorithms used to obtain these numbers, they are often inaccurate. .Pp The fields output are the user's login name, the name of the terminal the user is on, the host from which the user is logged in, the time the user
Re: strange load values
On Thu, 12 Mar 2009, jmc wrote: i believe that Theo and many of the devs have said many times that the load average means nothing. here's a reference to one such thread: http://kerneltrap.org/mailarchive/openbsd-misc/2008/11/8/4041294 2009/3/13 David Vasek va...@fido.cz: Perhaps the man page of w(1) should be changed. Currently it states: The load average numbers give the number of jobs in the run queue averaged over 1, 5 and 15 minutes. which is rather inaccurate or wrong. I suggest the patch below. Please correct my English if necessary. Regards, David Index: src/usr.bin/w/w.1 === RCS file: /cvs/src/usr.bin/w/w.1,v retrieving revision 1.18 diff -u -p -r1.18 w.1 --- src/usr.bin/w/w.1 31 May 2007 19:20:19 - 1.18 +++ src/usr.bin/w/w.1 13 Mar 2009 13:10:56 - @@ -50,7 +50,8 @@ The first line displays the current time been running, the number of users logged into the system, and the load averages. The load average numbers give the number of jobs in the run queue averaged -over 1, 5 and 15 minutes. +over past 1, 5 and 15 minutes. Because of the sampling method and +algorithms used to obtain these numbers, they are often inaccurate. .Pp At the peril of being justifiably told that I don't really know what I'm talking about here: I have a hunch that that last sentence may prevent that diff from being accepted. The way I understand things, the numbers are not inaccurate just different from what Linux users might expect them to be. But who ever said that Linux was the measure of all things? Again, I don't really know what I'm talking about here, but that's my understanding of the matter. If it's incorrect, then sorry for the noise. regards, --ropers
Re: strange load values
On Fri, Mar 13, 2009 at 9:13 AM, David Vasek va...@fido.cz wrote: Perhaps the man page of w(1) should be changed. Currently it states: The load average numbers give the number of jobs in the run queue averaged over 1, 5 and 15 minutes. which is rather inaccurate or wrong. I suggest the patch below. Please correct my English if necessary. why is it wrong? wouldn't a better patch be to fix it?
Re: strange load values
On Fri, 13 Mar 2009, Ted Unangst wrote: On Fri, Mar 13, 2009 at 9:13 AM, David Vasek va...@fido.cz wrote: Perhaps the man page of w(1) should be changed. Currently it states: The load average numbers give the number of jobs in the run queue averaged over 1, 5 and 15 minutes. which is rather inaccurate or wrong. I suggest the patch below. Please correct my English if necessary. why is it wrong? wouldn't a better patch be to fix it? It is wrong becasue the computed numbers can be different from what is written in the specification (the man pages). The computed load average can be high on an almost idle machine and vice-versa. As is described here: http://marc.info/?l=openbsd-miscm=118703405121404 http://marc.info/?l=openbsd-miscm=93551115818166 http://marc.info/?l=openbsd-miscm=121849543013236 I guess that this is the cause for all the repeated discussions about high load average. It can't be fixed without redesigning large portion of the kernel, if it can be fixed at all and it would definetely be for some performance trade-off. Regards, David
Re: strange load values
On Fri, Mar 13, 2009 at 4:13 PM, David Vasek va...@fido.cz wrote: It is wrong becasue the computed numbers can be different from what is written in the specification (the man pages). The computed load average can be high on an almost idle machine and vice-versa. As is described here: http://marc.info/?l=openbsd-miscm=118703405121404 http://marc.info/?l=openbsd-miscm=93551115818166 http://marc.info/?l=openbsd-miscm=121849543013236 I guess that this is the cause for all the repeated discussions about high load average. It can't be fixed without redesigning large portion of the kernel, if it can be fixed at all and it would definetely be for some performance trade-off. In theory, it could be off. But none of the people complaining about it have yet demonstrated that it actually was. If the load is the average number of processes in the run queue, and you don't think that your load could possibly be as high as 1.0, then show us evidence that demonstrates how many jobs were really in the run queue. Calculate the right answer. Nobody does this. They only say the load is wrong, but they don't say what it should be (other than small). No partial credit unless you show your work. Until somebody puts up some numbers, it's silly to add statements like this number is inaccurate to the man page.
Re: strange load values
--- Gabri Mate [Wed, Mar 11, 2009 at 08:58:15PM +0100]: --- Hey there! I have a Compaq DL580 G1 with 4x700Mhz PIII, 2GB RAM, 2x36GB U320 SCSI on a HP SmartArray 5300 with 64MB BBU. It's a general purpose 'hobby' server. The average concurrent i believe that Theo and many of the devs have said many times that the load average means nothing. here's a reference to one such thread: http://kerneltrap.org/mailarchive/openbsd-misc/2008/11/8/4041294
Re: strange load values
On 10:26 Thu 12 Mar , jmc wrote: --- Gabri Mate [Wed, Mar 11, 2009 at 08:58:15PM +0100]: --- Hey there! I have a Compaq DL580 G1 with 4x700Mhz PIII, 2GB RAM, 2x36GB U320 SCSI on a HP SmartArray 5300 with 64MB BBU. It's a general purpose 'hobby' server. The average concurrent i believe that Theo and many of the devs have said many times that the load average means nothing. here's a reference to one such thread: http://kerneltrap.org/mailarchive/openbsd-misc/2008/11/8/4041294 Yep, soon after my message someone showed me the original thread. Sorry for that i haven't found it before. -- Gabri Mate
Re: strange load values
I can see sometimes load about 20 to 100 on some servers at work. Then they are really slow ;-) 2009/3/12 Gabri Mate mailingl...@modernbiztonsag.org: On 10:26 Thu 12 Mar B B , jmc wrote: --- Gabri Mate [Wed, Mar 11, 2009 at 08:58:15PM +0100]: --- Hey there! I have a Compaq DL580 G1 with 4x700Mhz PIII, 2GB RAM, 2x36GB U320 SCSI on a HP SmartArray 5300 with 64MB BBU. It's a general purpose 'hobby' server. The average concurrent i believe that Theo and many of the devs have said many times that the load average means nothing. here's a reference to one such thread: http://kerneltrap.org/mailarchive/openbsd-misc/2008/11/8/4041294 Yep, soon after my message someone showed me the original thread. Sorry for that i haven't found it before. -- Gabri Mate -- http://www.openbsd.org/lyrics.html