[PERFORM] shared_buffers on ubuntu precise
On Nov 30, 2012, at 8:06 AM, Shaun Thomas wrote: I say that because you mentioned you're using Ubuntu 12.04, and we were having some problems with PG on that platform. With shared_buffers over 4GB, it starts doing really weird things to the memory subsystem. Whatever it does causes the kernel to purge cache rather aggressively. We saw a 60% reduction in read IO by reducing shared_buffers to 4GB. Without as many reads, your writes should be much less disruptive. Hm, this sounds like something we should look into. Before we start digging do you have more to share, or did you leave it with the huh, that's weird; this seems to fix it solution?
Re: [PERFORM] shared_buffers on ubuntu precise
On 11/30/2012 01:57 PM, Ben Chobot wrote: Hm, this sounds like something we should look into. Before we start digging do you have more to share, or did you leave it with the huh, that's weird; this seems to fix it solution? We're still testing. We're still on the -31 kernel. We tried the -33 kernel which *might* fix it, but then this happened: https://bugs.launchpad.net/ubuntu/+source/linux/+bug/1084264 So now we're testing -34 which is currently proposed. Either way, it's pretty clear that Ubuntu's choice of patches to backport is rather eclectic and a little wonky, or that nailing down load calculations went awry since the NOHZ stuff started, or both. At this point, I wish we'd stayed on CentOS. -- Shaun Thomas OptionsHouse | 141 W. Jackson Blvd. | Suite 500 | Chicago IL, 60604 312-444-8534 stho...@optionshouse.com __ See http://www.peak6.com/email_disclaimer/ for terms and conditions related to this email -- Sent via pgsql-performance mailing list (pgsql-performance@postgresql.org) To make changes to your subscription: http://www.postgresql.org/mailpref/pgsql-performance
Re: [PERFORM] shared_buffers on ubuntu precise
On Fri, Nov 30, 2012 at 02:01:45PM -0600, Shaun Thomas wrote: On 11/30/2012 01:57 PM, Ben Chobot wrote: Hm, this sounds like something we should look into. Before we start digging do you have more to share, or did you leave it with the huh, that's weird; this seems to fix it solution? We're still testing. We're still on the -31 kernel. We tried the -33 kernel which *might* fix it, but then this happened: https://bugs.launchpad.net/ubuntu/+source/linux/+bug/1084264 So now we're testing -34 which is currently proposed. Either way, it's pretty clear that Ubuntu's choice of patches to backport is rather eclectic and a little wonky, or that nailing down load calculations went awry since the NOHZ stuff started, or both. At this point, I wish we'd stayed on CentOS. Or Debian. Not sure what would justify use of Ubuntu as a server, except wanting to have the exact same OS as their personal computers. -- Bruce Momjian br...@momjian.ushttp://momjian.us EnterpriseDB http://enterprisedb.com + It's impossible for everything to be true. + -- Sent via pgsql-performance mailing list (pgsql-performance@postgresql.org) To make changes to your subscription: http://www.postgresql.org/mailpref/pgsql-performance
Re: [PERFORM] shared_buffers on ubuntu precise
On 11/30/2012 02:38 PM, Bruce Momjian wrote: Or Debian. Not sure what would justify use of Ubuntu as a server, except wanting to have the exact same OS as their personal computers. Honestly not sure why we went that direction. I'm not in the sysadmin group, though I do work with them pretty closely. I think it was because of the LTS label, and the fact that the packages are quite a bit more recent than Debian stable. I can say however, that I'm testing the 3.4 kernel right now, and it seems much better. I may be able to convince them to install that instead if their own tests prove beneficial. -- Shaun Thomas OptionsHouse | 141 W. Jackson Blvd. | Suite 500 | Chicago IL, 60604 312-444-8534 stho...@optionshouse.com __ See http://www.peak6.com/email_disclaimer/ for terms and conditions related to this email -- Sent via pgsql-performance mailing list (pgsql-performance@postgresql.org) To make changes to your subscription: http://www.postgresql.org/mailpref/pgsql-performance
Re: [PERFORM] shared_buffers on ubuntu precise
On Fri, Nov 30, 2012 at 12:38 PM, Bruce Momjian br...@momjian.us wrote: Or Debian. Not sure what would justify use of Ubuntu as a server, except wanting to have the exact same OS as their personal computers. We have switched from Debian to Ubuntu: there is definitely non-zero value in the PPA hosting (although it's rather terrible in many ways), regular LTS releases (even if you choose not to use them right away, and know they are somewhat buggy at times), and working with AWS and Canonical as organizations (that, most importantly, can interact directly without my own organization) on certain issues. For example, this dog of a bug: https://bugs.launchpad.net/ubuntu/+source/linux-ec2/+bug/929941 I also frequently take advantage of Debian unstable for backporting of specific packages that are very important to me, so there's a lot of value to me in Ubuntu being quite similar to Debian. In fact, even though I say we 'switched', it's not as though we re-did some entrenched systems from Debian to Ubuntu -- rather, we employ both systems at the same time and I don't recall gnashing of teeth about that, because they are very similar. Yet, there is a clear Ubuntu preference for new systems made today and, to wit, I can't think of anyone with more than the most mild preference for Debian. Conversely, I'd say the preference for Ubuntu for the aforementioned reasons is clear but moderate at most. Also, there's the similarity to the lap/desktop environment. Often cited with some derision, yet it does add a lot of value, even if people run slightly newer Ubuntus on their non-production computer. -- fdr -- Sent via pgsql-performance mailing list (pgsql-performance@postgresql.org) To make changes to your subscription: http://www.postgresql.org/mailpref/pgsql-performance
Re: [PERFORM] shared_buffers on ubuntu precise
On 01/12/12 11:21, Daniel Farina wrote: On Fri, Nov 30, 2012 at 12:38 PM, Bruce Momjian br...@momjian.us wrote: Or Debian. Not sure what would justify use of Ubuntu as a server, except wanting to have the exact same OS as their personal computers. We have switched from Debian to Ubuntu: there is definitely non-zero value in the PPA hosting (although it's rather terrible in many ways), regular LTS releases (even if you choose not to use them right away, and know they are somewhat buggy at times), and working with AWS and Canonical as organizations (that, most importantly, can interact directly without my own organization) on certain issues. For example, this dog of a bug: https://bugs.launchpad.net/ubuntu/+source/linux-ec2/+bug/929941 I also frequently take advantage of Debian unstable for backporting of specific packages that are very important to me, so there's a lot of value to me in Ubuntu being quite similar to Debian. In fact, even though I say we 'switched', it's not as though we re-did some entrenched systems from Debian to Ubuntu -- rather, we employ both systems at the same time and I don't recall gnashing of teeth about that, because they are very similar. Yet, there is a clear Ubuntu preference for new systems made today and, to wit, I can't think of anyone with more than the most mild preference for Debian. Conversely, I'd say the preference for Ubuntu for the aforementioned reasons is clear but moderate at most. Also, there's the similarity to the lap/desktop environment. Often cited with some derision, yet it does add a lot of value, even if people run slightly newer Ubuntus on their non-production computer. +1 We have gone through pretty much the same process in the last couple of years. Most of our new systems run Ubuntu, some Debian. There is definitely value in running the same system on the desktop too - often makes bug replication ridiculously easy (no having to find the appropriate test environment, ask if I can hammer/punish/modify it etc etc, and no need even spin up a VM). Cheers Mark -- Sent via pgsql-performance mailing list (pgsql-performance@postgresql.org) To make changes to your subscription: http://www.postgresql.org/mailpref/pgsql-performance
