Date: Mon, 10 Apr 2006 19:20:33 +0200
To: "Jesper Krogh" <jesper@krogh.cc>, pgsql-performance@postgresql.org
Subject: Re: Restore performance?
References: <e1d31l$2nt$1@sea.gmane.org>
From: PFC <lists@peufeu.com>
Content-Type: text/plain; format=flowed; delsp=yes; charset=utf-8
MIME-Version: 1.0
Content-Transfer-Encoding: 8bit
Message-ID: <op.s7sygjtecigqcu@apollo13>
In-Reply-To: <e1d31l$2nt$1@sea.gmane.org>
User-Agent: Opera M2/8.51 (Linux, build 1462)


> I'd run pg_dump | gzip > sqldump.gz  on the old system.

	If the source and destination databases are on different machines, you  
can pipe pg_dump on the source machine to pg_restore on the destination  
machine by using netcat.
	If you only have 100 Mbps ethernet, compressing the data will be faster.  
If you have Gb Ethernet, maybe you don't need to compress, but it doesn't  
hurt to test.

	use pg_restore instead of psql, and use a recent version of pg_dump which  
can generate dumps in the latest format.

	If you need fast compression, use gzip -1 or even lzop, which is  
incredibly fast.

	Turn off fsync during the restore and set maintenance_work_mem to use  
most of your available RAM for index creation.

	I think that creating foreign key constraints uses large joins ; it might  
be good to up work_mem also.

	Check the speed of your disks with dd beforehand. You might get a  
surprise.

	Maybe you can also play with the bgwriter and checkpoint parameters.