pg.ddx.io  pgsql-performance@postgresql.org mailing list archive  
help / color / mirror / Atom feed
From: Shridhar Daithankar <shridhar@frodo.hserus.net>
To: Missner, T. R. <T.R.Missner@Level3.com>
Cc: pgsql-performance@postgresql.org
Subject: Re: inserting into brand new database faster than old database
Date: Thu, 08 Jul 2004 14:32:43 +0530
Message-ID: <40ED0DB3.3080209@frodo.hserus.net> (raw)
In-Reply-To: <F67BE6C5DEBB8340A749DE01B476FEC103E4D3BC@idc1exc0002.corp.global.level3.com>
References: <F67BE6C5DEBB8340A749DE01B476FEC103E4D3BC@idc1exc0002.corp.global.level3.com>

Missner, T. R. wrote:

> Hello,
> 
> I have been a happy postgresql developer for a few years now.  Recently
> I have discovered a very strange phenomenon in regards to inserting
> rows.
> 
> My app inserts millions of records a day, averaging about 30 rows a
> second. I use autovac to make sure my stats and indexes are up to date.
> Rarely are rows ever deleted.  Each day a brand new set of tables is
> created and eventually the old tables are dropped. The app calls
> functions which based on some simple logic perform the correct inserts.

Have you profiled where the time goes in a brand new schema and a degraded 
database? Is it IO? Is it CPU? Is the function making decision becoming bottleneck?

> The problem I am seeing is that after a particular database gets kinda
> old, say a couple of months, performance begins to degrade.  Even after
> creating brand new tables my insert speed is slow in comparison ( by a
> magnitude of 5 or more ) with a brand new schema which has the exact
> same tables.  I am running on an IBM 360 dual processor Linux server
> with a 100 gig raid array spanning 5 scsi disks.  The machine has 1 gig
> of ram of which 500 meg is dedicated to Postgresql.
> 
> Just to be clear, the question I have is why would a brand new db schema
> allow inserts faster than an older schema with brand new tables?  Since
> the tables are empty to start, vacuuming should not be an issue at all.
> Each schema is identical in every way except the db name and creation
> date.

You can do few things.

- Get explain analyze. See the difference between actual and projected timings. 
The difference is the hint about where planner is going wrong.

- Is IO your bottleneck? Are vacuum taking longer and longer? If yes then you 
could try the vacuum delay patch. If your IO is saturated for any reason, 
everything is going to crawl

- Are your indexes bloat free? If you are using pre7.x,vacuum does not clean up 
indexes. You need to reindex.

- Have you vacuumed the complete database? If the catalogs collect dead space it 
could cause degradation too but that is just a guess.

Basically monitor slow inserts and try to find out where time is spent.

HTH

  Shridhar



view thread (5+ messages)  latest in thread

Message-ID: <40ED0DB3.3080209@frodo.hserus.net>
Permalink:  ../40ED0DB3.3080209@frodo.hserus.net/
Also on:    postgresql.org/message-id/40ED0DB3.3080209@frodo.hserus.net

 · 

reply

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Reply to all the recipients using the --to and --cc options:
  reply via email

  To: pgsql-performance@postgresql.org
  Cc: shridhar@frodo.hserus.net, T.R.Missner@Level3.com
  Subject: Re: inserting into brand new database faster than old database
  In-Reply-To: <40ED0DB3.3080209@frodo.hserus.net>

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

This inbox is served by DDX for PostgreSQL; see mirroring instructions
for how to clone and mirror all data and code used for this inbox