From: | Tom Lane <tgl(at)sss(dot)pgh(dot)pa(dot)us> |
---|---|
To: | Ben Clewett <B(dot)Clewett(at)roadrunner(dot)uk(dot)com> |
Cc: | "Matthew T(dot) O'Connor" <matthew(at)zeut(dot)net>, pgsql-hackers(at)postgresql(dot)org |
Subject: | Re: For the ametures. (related to "Are we losing momentum?") |
Date: | 2003-04-22 15:18:45 |
Message-ID: | 3032.1051024725@sss.pgh.pa.us |
Views: | Raw Message | Whole Thread | Download mbox | Resend email |
Thread: | |
Lists: | pgsql-hackers |
Ben Clewett <B(dot)Clewett(at)roadrunner(dot)uk(dot)com> writes:
> If I wanted to divide the postmaster read() calls evenly to files
> located over several physical disks, how would you suggest distributing
> the data-space?
AFAIK, the single biggest win you can get in this dimension is to put
the WAL log ($PGDATA/pg_xlog/) on a separate spindle from everything
else. At least for write-intensive databases, that can buy you
something like 2x improvement for the price of one easy symlink.
After that, the conventional wisdom is to put indexes on a third spindle
(separate from base tables and from xlog). But the bookkeeping and
maintenance effort needed for that is really too high to make it worth
worrying about, IMHO :-(. Eventually we will have some kind of
tablespace feature to make it easy.
My recommendation at the moment would be: WAL on dedicated spindle,
everything else on the best RAID array you can set up. And buy as much
RAM as you can afford.
See past discussions in pgsql-performance for more info.
regards, tom lane
From | Date | Subject | |
---|---|---|---|
Next Message | Tom Lane | 2003-04-22 15:31:43 | Re: bit strings - anyone working on them? |
Previous Message | Tilo Schwarz | 2003-04-22 15:14:07 | Re: Are we losing momentum? |