Re: [HACKERS] Fix performance degradation of contended LWLock on NUMA

On 2017-10-20 11:54, Sokolov Yura wrote: Hello, On 2017-10-19 19:46, Andres Freund wrote: On 2017-10-19 14:36:56 +0300, Sokolov Yura wrote: > > + init_local_spin_delay(); > > The way you moved this around has the disadvantage that we now do this - > a number of writes - even in the very

Re: [HACKERS] Fix performance degradation of contended LWLock on NUMA

Hello, On 2017-10-19 19:46, Andres Freund wrote: On 2017-10-19 14:36:56 +0300, Sokolov Yura wrote: > > + init_local_spin_delay(); > > The way you moved this around has the disadvantage that we now do this - > a number of writes - even in the very common case where the lwlock can > be

Re: [HACKERS] Fix performance degradation of contended LWLock on NUMA

On 2017-10-19 14:36:56 +0300, Sokolov Yura wrote: > > > + init_local_spin_delay(); > > > > The way you moved this around has the disadvantage that we now do this - > > a number of writes - even in the very common case where the lwlock can > > be acquired directly. > > Excuse me, I don't

Re: [HACKERS] Fix performance degradation of contended LWLock on NUMA

On 2017-10-19 02:28, Andres Freund wrote: On 2017-06-05 16:22:58 +0300, Sokolov Yura wrote: Algorithm for LWLockWaitForVar is also refactored. New version is: 1. If lock is not held by anyone, it immediately exit. 2. Otherwise it is checked for ability to take WaitList lock, because variable

Re: [HACKERS] Fix performance degradation of contended LWLock on NUMA

Hi, On 2017-10-19 03:03, Andres Freund wrote: Hi, On 2017-09-08 22:35:39 +0300, Sokolov Yura wrote: /* * Internal function that tries to atomically acquire the lwlock in the passed - * in mode. + * in mode. If it could not grab the lock, it doesn't puts proc into wait + * queue. * -

Re: [HACKERS] Fix performance degradation of contended LWLock on NUMA

Hi, On 2017-09-08 22:35:39 +0300, Sokolov Yura wrote: > From 722a8bed17f82738135264212dde7170b8c0f397 Mon Sep 17 00:00:00 2001 > From: Sokolov Yura > Date: Mon, 29 May 2017 09:25:41 + > Subject: [PATCH 1/6] More correct use of spinlock inside LWLockWaitListLock.

Re: [HACKERS] Fix performance degradation of contended LWLock on NUMA

Hi, On 2017-06-05 16:22:58 +0300, Sokolov Yura wrote: > Patch changes the way LWLock is acquired. > > Before patch, lock is acquired using following procedure: > 1. first LWLock->state is checked for ability to take a lock, and CAS > performed (either lock could be acquired or not). And it is

Re: [HACKERS] Fix performance degradation of contended LWLock on NUMA

On 09/11/2017 11:01 AM, Jesper Pedersen wrote: Thanks for working on this ! Moved to "Ready for Committer". Best regards,  Jesper -- Sent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org) To make changes to your subscription: http://www.postgresql.org/mailpref/pgsql-hackers

Re: [HACKERS] Fix performance degradation of contended LWLock on NUMA

Hi, On 09/08/2017 03:35 PM, Sokolov Yura wrote: I'm seeing -M prepared: Up to 11% improvement -M prepared -S: No improvement, no regression ("noise") -M prepared -N: Up to 12% improvement for all runs the improvement shows up the closer you get to the number of CPU threads, or above. Although

Re: [HACKERS] Fix performance degradation of contended LWLock on NUMA

Hi, Jesper Thank you for reviewing. On 2017-09-08 18:33, Jesper Pedersen wrote: Hi, On 07/18/2017 01:20 PM, Sokolov Yura wrote: I'm sending rebased version with couple of one-line tweaks. (less skip_wait_list on shared lock, and don't update spin-stat on aquiring) I have been running

Re: [HACKERS] Fix performance degradation of contended LWLock on NUMA

Hi, On 07/18/2017 01:20 PM, Sokolov Yura wrote: I'm sending rebased version with couple of one-line tweaks. (less skip_wait_list on shared lock, and don't update spin-stat on aquiring) I have been running this patch on a 2S/28C/56T/256Gb w/ 2 x RAID10 SSD setup (1 to 375 clients on logged

Re: [HACKERS] Fix performance degradation of contended LWLock on NUMA

On 2017-07-18 20:20, Sokolov Yura wrote: On 2017-06-05 16:22, Sokolov Yura wrote: Good day, everyone. This patch improves performance of contended LWLock. Patch makes lock acquiring in single CAS loop: 1. LWLock->state is read, and ability for lock acquiring is detected. If there is

Re: [HACKERS] Fix performance degradation of contended LWLock on NUMA

On 2017-06-05 16:22, Sokolov Yura wrote: Good day, everyone. This patch improves performance of contended LWLock. Patch makes lock acquiring in single CAS loop: 1. LWLock->state is read, and ability for lock acquiring is detected. If there is possibility to take a lock, CAS tried. If CAS

Re: [HACKERS] Fix performance degradation of contended LWLock on NUMA

On Mon, Jun 5, 2017 at 9:22 AM, Sokolov Yura wrote: > Good day, everyone. > > This patch improves performance of contended LWLock. > It was tested on 4 socket 72 core x86 server (144 HT) Centos 7.1 > gcc 4.8.5 > > Patch makes lock acquiring in single CAS loop: > 1.

[HACKERS] Fix performance degradation of contended LWLock on NUMA

Good day, everyone. This patch improves performance of contended LWLock. It was tested on 4 socket 72 core x86 server (144 HT) Centos 7.1 gcc 4.8.5 Results: pgbench -i -s 300 + pgbench --skip-some-updates Clients | master | patched +=+=== 30 |32k |32k 50