On 19.11.2012 г. 16:59 ч., Poul Dige wrote:
Fra: Alex Peshkoff [mailto:peshk...@mail.ru]On 11/13/12 18:18, Poul Dige wrote:I could put a screen dump here of the Windows 2008R2/63 task managerwith AMD Opteron 6274, 2x16 core running FB2.5.1 SC/32 bit (due to some 32 bit UDF). I don't know if you are interested at a gaze. We see exactly the same kind of usage, 8 cores are in use and 24 are more orless doing nothing.However, the cores in use are not maxed out - so it COULD be somethingwith power management that the OS doesn't want to activate more CPU's than necessary. I can't tell about that for sure. Is it SC or CS?Hi Alex, it is Super Classic.Let me pay your attention that in initially described case use of classic made all cores loaded. How does your server behave with classic?Hi Alex, It is a very important production server and I can't find a good way to make tests on it by switching to classic. Thing is, the connections are very short lived, so it is a nice feature that the threads are spawn very quickly by Firebird. I have seen another, similar server (32 cores Win2k8R2) with FB2.1 CS which seems to distribute load evenly, so I guess the problem lies in thread handling - not if FB but in Windows itself. I will try to install an update for Windows as it seems to be a well known problem with Windows 7/2008R2 vs AMD Bulldozer architecture, so the following hotfix may just be a solution (there are two hotfixes, the other one is referred to in the article): http://support.microsoft.com/kb/2646060 I will re-post once we have had opportunity to update the server, which will probably not be until the weekend.I installed the above mentioned hotfix from MS (toghether with the prerequisit KB2645594) this past weekend. Today I see: - exactly the same. 55 threads (according to task manager) in FB 2.5.1.26351 SC (Super Classic) spread over only 8 of the 32 cores on the Win2008R2 SP1 server (see hw specs above). The cores aren't max'ed out, so at least in princip there is an excuse for not firing up the remaining 24 cores. But, somehow I'd be more happy to see an equal load on all cores. In other words - the hotfix didn't change anything noticably. Does FB decide anything regarding on which cores to run, or is it entirely a Windows decision? I imagine the latter, but of course we have the "affinity" parameter in the config file for super server installations, so at least it is possible to influence choice of processor core. Are there any suggestions what I should try to do next? We have actually 2 similar servers, both running 2.5.1SC (for now). The other server COULD be equipped with CS instead, if that would make a lot of sence as a trial. Even SS, as it primarily serves a lot of different databases with only a few connections on each. Your call, Alex :) Best regards Poul
Could be the fact that this server uses NUMA architecture and windows is trying to keep the process in the same memory bank where it was initially started in order to avoid expensive HT transactions to reach to other memory banks connected on other CPU cores.
With CS you will have separate process and in this case windows can distribute each process on separate core and memory is per process.
-- Doychin Bondzhev dSoft-Bulgaria Ltd. PowerPro - billing & provisioning solution for Service providers PowerStor - Warehouse & POS http://www.dsoft-bg.com/ Mobile: +359888243116
<<attachment: doychin.vcf>>
------------------------------------------------------------------------------ Monitor your physical, virtual and cloud infrastructure from a single web console. Get in-depth insight into apps, servers, databases, vmware, SAP, cloud infrastructure, etc. Download 30-day Free Trial. Pricing starts from $795 for 25 servers or applications! http://p.sf.net/sfu/zoho_dev2dev_nov
Firebird-Devel mailing list, web interface at https://lists.sourceforge.net/lists/listinfo/firebird-devel
