5 ms·

Good question. It was mentioned in the talk (http://www.slideshare.net/fuzzycz/performance-archaeology-40583681 http://www.slideshare.net/fuzzycz/performance-ar

Good question. It was mentioned in the talk (http://www.slideshare.net/fuzzycz/performance-archaeology-40583681 http://www.slideshare.net/fuzzycz/performance-archaeology-40...), but I forgot to put that into the post. Both machines were running Linux. The DL380 machine (the 'main' one) was running Scientific Linux 6.5 with 2.6.32 kernel and XFS. The i5-2500k machine was running Gentoo (so mostly 'bleeding edge' packages), with kernel 3.12 and ext4. Update: Actually, I've just noticed it's mentioned in the previous 'introduction' post in the series (http://blog.pgaddict.com/posts/performance-since-postgresql-7-4-to-9-4 http://blog.pgaddict.com/posts/performance-since-postgresql-...).

sophacles 12y ago
Does the difference in kernel an filesystem have any effect on the jumps you saw between the xeon and i5? Really curious about how the benchmarks comparing the two play out if the hardwares were both running Gentoo with 3.12 and ext4. (or the older kernel and xfs I guess...).
- pgaddict 12y ago
  Good point. For example the 9.2 changes were done after in-kernel lseek() optimizations, which improved the performance in some workloads. So yes, that might be a factor here (and I should have point that out in the post). I can't easily reinstall the OS on the Xeon machine (it's in a server room that I don't have direct access to), but installing an older kernel on the i5 machine should not be a problem. Hmmm, a nice topic for another benchmarking quest ;-)
mrmondo 12y ago
Kernel 2.6.32 on that DL380 is ancient regardless of whatever patches are shoehorned in by Redhat it's very old tech - you'd notice considerably better performance when using a modern kernel. We also noticed a huge gain in performance from upgrading from Kernel 3.14 to 3.18 especially with regards to storage and network latency.
- pgaddict 12y ago
  Well, you're right. The thing is - it's the kernel version for that particular distribution, tracking RH 6. That's one of the reasons why I did the tests on the Gentoo box with a much newer kernel.
  - mrmondo 12y ago
    Yeah the ecosystem for Redhat and variants becomes legacy very quickly - don't get me wrong Redhat are an amazing company and make quality software but when you need performance or newer features they aren't the best. Debian wheezy has had 3.16 stable in Backports since its release as that's what Debain Jessie will be released with shortly. We (Infoxchange) actually CI our kernel builds daily so we end up with whatever the latest stable kernel is as its released, we also use that as an oppertunity to add the GRSecurity patches to the kernel.
    - pgaddict 12y ago
      Different distributions for different users/customers, I guess. People who run RH often require stability over long periods of time, but if you want faster update cycle, RH (or derived distros) might not be the right choices. Hmmm, but maybe I could use the kernel from epel. I see there's 3.10 (lt), 3.18 and 3.19 (ml) for el6. It's a bit tricky though, because I have absolutely no access to this machine except for ssh (no KVM or anything), so a botched kernel upgrade means a few days of outage before it's taken care of (it's hosted at a local university for free).
      - mrmondo 12y ago
        Generally speaking newer kernels are considerably more stable than older kernels - the main reason Redhat and variants have such old Kernels is because of the proprietary driver compatibility and patches for 'enterprise' hardware such as RAID controllers. Additionally - there is also a trap where software is deployed that has been developed without the idea of being modernised (and in some cases maintained) over time - Vendors will often charge large amounts of money to modernise and test these applications. I've observed this as both a means of increasing profitability, out of lack of quality engineering / development skills or practices or by misinforming customers to think that risk aversion and a slow rate of change is the least costly option over time which leads to or is caused by pathological or bureaucratic organisational culture.

sophacles 12y ago

Does the difference in kernel an filesystem have any effect on the jumps you saw between the xeon and i5? Really curious about how the benchmarks comparing the two play out if the hardwares were both running Gentoo with 3.12 and ext4. (or the older kernel and xfs I guess...).

pgaddict 12y ago
Good point. For example the 9.2 changes were done after in-kernel lseek() optimizations, which improved the performance in some workloads. So yes, that might be a factor here (and I should have point that out in the post). I can't easily reinstall the OS on the Xeon machine (it's in a server room that I don't have direct access to), but installing an older kernel on the i5 machine should not be a problem. Hmmm, a nice topic for another benchmarking quest ;-)

mrmondo 12y ago

Kernel 2.6.32 on that DL380 is ancient regardless of whatever patches are shoehorned in by Redhat it's very old tech - you'd notice considerably better performance when using a modern kernel. We also noticed a huge gain in performance from upgrading from Kernel 3.14 to 3.18 especially with regards to storage and network latency.

pgaddict 12y ago
Well, you're right. The thing is - it's the kernel version for that particular distribution, tracking RH 6. That's one of the reasons why I did the tests on the Gentoo box with a much newer kernel.
- mrmondo 12y ago
  Yeah the ecosystem for Redhat and variants becomes legacy very quickly - don't get me wrong Redhat are an amazing company and make quality software but when you need performance or newer features they aren't the best. Debian wheezy has had 3.16 stable in Backports since its release as that's what Debain Jessie will be released with shortly. We (Infoxchange) actually CI our kernel builds daily so we end up with whatever the latest stable kernel is as its released, we also use that as an oppertunity to add the GRSecurity patches to the kernel.
  - pgaddict 12y ago
    Different distributions for different users/customers, I guess. People who run RH often require stability over long periods of time, but if you want faster update cycle, RH (or derived distros) might not be the right choices. Hmmm, but maybe I could use the kernel from epel. I see there's 3.10 (lt), 3.18 and 3.19 (ml) for el6. It's a bit tricky though, because I have absolutely no access to this machine except for ssh (no KVM or anything), so a botched kernel upgrade means a few days of outage before it's taken care of (it's hosted at a local university for free).
    - mrmondo 12y ago
      Generally speaking newer kernels are considerably more stable than older kernels - the main reason Redhat and variants have such old Kernels is because of the proprietary driver compatibility and patches for 'enterprise' hardware such as RAID controllers. Additionally - there is also a trap where software is deployed that has been developed without the idea of being modernised (and in some cases maintained) over time - Vendors will often charge large amounts of money to modernise and test these applications. I've observed this as both a means of increasing profitability, out of lack of quality engineering / development skills or practices or by misinforming customers to think that risk aversion and a slow rate of change is the least costly option over time which leads to or is caused by pathological or bureaucratic organisational culture.