From: | Mithun Cy <mithun(dot)cy(at)enterprisedb(dot)com> |
---|---|
To: | Amit Kapila <amit(dot)kapila16(at)gmail(dot)com> |
Cc: | Robert Haas <robertmhaas(at)gmail(dot)com>, Andres Freund <andres(at)anarazel(dot)de>, pgsql-hackers <pgsql-hackers(at)postgresql(dot)org> |
Subject: | Re: POC: Cache data in GetSnapshotData() |
Date: | 2016-03-03 11:20:17 |
Message-ID: | CAD__OuiwEi5sHe2wwQCK36Ac9QMhvJuqG3CfPN+OFCMb7rdruQ@mail.gmail.com |
Views: | Raw Message | Whole Thread | Download mbox | Resend email |
Thread: | |
Lists: | pgsql-hackers |
On Tue, Mar 1, 2016 at 12:59 PM, Amit Kapila <amit(dot)kapila16(at)gmail(dot)com>
wrote:
>Don't we need to add this only when the xid of current transaction is
valid? Also, I think it will be better if we can explain why we need to
add the our >own transaction id while caching the snapshot.
I have fixed the same thing and patch is attached.
Some more tests done after that
*pgbench write tests: on 8 socket, 64 core machine.*
/postgres -c shared_buffers=16GB -N 200 -c min_wal_size=15GB -c
max_wal_size=20GB -c checkpoint_timeout=900 -c maintenance_work_mem=1GB -c
checkpoint_completion_target=0.9
./pgbench -c $clients -j $clients -T 1800 -M prepared postgres
[image: Inline image 3]
A small improvement in performance at 64 thread.
*LWLock_Stats data:*
ProcArrayLock: Base.
=================
postgresql-2016-03-01_115252.log:PID 110019 lwlock main 4: shacq 1867601
exacq 35625 blk 134682 spindelay 128 dequeue self 28871
postgresql-2016-03-01_115253.log:PID 110115 lwlock main 4: shacq 2201613
exacq 43489 blk 155499 spindelay 127 dequeue self 32751
postgresql-2016-03-01_115253.log:PID 110122 lwlock main 4: shacq 2231327
exacq 44824 blk 159440 spindelay 128 dequeue self 33336
postgresql-2016-03-01_115254.log:PID 110126 lwlock main 4: shacq 2247983
exacq 44632 blk 158669 spindelay 131 dequeue self 33365
postgresql-2016-03-01_115254.log:PID 110059 lwlock main 4: shacq 2036809
exacq 38607 blk 143538 spindelay 117 dequeue self 31008
ProcArrayLock: With Patch.
=====================
postgresql-2016-03-01_124747.log:PID 1789 lwlock main 4: shacq 2273958
exacq 61605 blk 79581 spindelay 307 dequeue self 66088
postgresql-2016-03-01_124748.log:PID 1880 lwlock main 4: shacq 2456388
exacq 65996 blk 82300 spindelay 470 dequeue self 68770
postgresql-2016-03-01_124748.log:PID 1765 lwlock main 4: shacq 2244083
exacq 60835 blk 79042 spindelay 336 dequeue self 65212
postgresql-2016-03-01_124749.log:PID 1882 lwlock main 4: shacq 2489271
exacq 67043 blk 85650 spindelay 463 dequeue self 68401
postgresql-2016-03-01_124749.log:PID 1753 lwlock main 4: shacq 2232791
exacq 60647 blk 78659 spindelay 364 dequeue self 65180
postgresql-2016-03-01_124750.log:PID 1849 lwlock main 4: shacq 2374922
exacq 64075 blk 81860 spindelay 339 dequeue self 67584
*-------------Block time of ProcArrayLock has reduced significantly.*
ClogControlLock : Base.
===================
postgresql-2016-03-01_115302.log:PID 110040 lwlock main 11: shacq 586129
exacq 268808 blk 90570 spindelay 261 dequeue self 59619
postgresql-2016-03-01_115303.log:PID 110047 lwlock main 11: shacq 593492
exacq 271019 blk 89547 spindelay 268 dequeue self 59999
postgresql-2016-03-01_115303.log:PID 110078 lwlock main 11: shacq 620830
exacq 285244 blk 92939 spindelay 262 dequeue self 61912
postgresql-2016-03-01_115304.log:PID 110083 lwlock main 11: shacq 633101
exacq 289983 blk 93485 spindelay 262 dequeue self 62394
postgresql-2016-03-01_115305.log:PID 110105 lwlock main 11: shacq 646584
exacq 297598 blk 93331 spindelay 312 dequeue self 63279
ClogControlLock : With Patch.
=======================
postgresql-2016-03-01_124730.log:PID 1865 lwlock main 11: shacq 722881
exacq 330273 blk 106163 spindelay 468 dequeue self 80316
postgresql-2016-03-01_124731.log:PID 1857 lwlock main 11: shacq 713720
exacq 327158 blk 106719 spindelay 439 dequeue self 79996
postgresql-2016-03-01_124732.log:PID 1826 lwlock main 11: shacq 696762
exacq 317239 blk 104523 spindelay 448 dequeue self 79374
postgresql-2016-03-01_124732.log:PID 1862 lwlock main 11: shacq 721272
exacq 330350 blk 105965 spindelay 492 dequeue self 81036
postgresql-2016-03-01_124733.log:PID 1879 lwlock main 11: shacq 737398
exacq 335357 blk 105424 spindelay 520 dequeue self 80977
*-------------Block time of ClogControlLock has increased slightly.*
Will continue with further tests on lower clients.
--
Thanks and Regards
Mithun C Y
EnterpriseDB: http://www.enterprisedb.com
Attachment | Content-Type | Size |
---|---|---|
Cache_data_in_GetSnapshotData_POC_01.patch | application/octet-stream | 8.4 KB |
From | Date | Subject | |
---|---|---|---|
Next Message | Michael Paquier | 2016-03-03 11:28:35 | Re: [NOVICE] WHERE clause not used when index is used |
Previous Message | Michael Paquier | 2016-03-03 11:19:18 | Re: Re: redo failed in physical streaming replication while stopping the master server |