Performance: fsync vs. commit delay

From: "Ilja Golshtein" <ilejn(at)yandex(dot)ru>
To: pgsql-general(at)postgresql(dot)org
Cc: teodor(at)sigaev(dot)ru, oleg(at)sai(dot)msu(dot)su
Subject: Performance: fsync vs. commit delay
Date: 2006-10-20 09:40:42
Message-ID: 4538999A.000003.18652@colgate.yandex.ru
Views: Raw Message | Whole Thread | Download mbox | Resend email
Thread:
Lists: pgsql-general

Hello!

I run an [almost] OLTP system with PG as backend on Windows and on Linux.

The idea is obvious - increase performance as much as possible keeping reliability in mind and I'd highly appreciate any hints.

Some questions about performance/reliability trade.

My first idea was to increase commit_delay having fsync on, since I believed commit_delay is rather conservative thing. Reading docs I realized it is actually not, because it prevents server from immediate writes. It means if I disable fsync only OS/hardware crashes are dangerous, while after setting nonzero commit_delay PG server crashes are harmful as well.

Is my understanding correct?

What I want is to do write always but do fsync, say, every 100 ms. Is it possible to control fsyncs and writes separately, e.g. have write for every commit and fsync as often as specified time range?

--
Best regards
Ilja Golshtein

Browse pgsql-general by date

  From Date Subject
Next Message Andreas Seltenreich 2006-10-20 10:27:38 Re: why not kill -9 postmaster
Previous Message Karl O. Pinc 2006-10-20 09:10:51 Re: plpgsql FOUND bug when SELECT INTO assigns a NULL