On Thu, Sep 16, 2004 at 10:50:33AM -0700, Josh Berkus wrote:
> The second step is to have lots of disks; 5 drives is a minimum for really
> good performance. 3-drive RAID5, in particular, is a poor performer for
> PostgreSQL, often resulting in I/O that is 40% or less as efficient as a
> single disk due to extremely slow random seeks and little parallelization.
> Once you have 6 drives or more, opinions are divided on whether RAID 10 or
> RAID 5 is better. I think it partly depends on your access pattern.
What about benefits from putting WAL and pg_temp on seperate drives?
Specifically, we have a box with 8 drives, 2 in a mirror with the OS and
WAL and pg_temp; the rest in a raid10 with the database on it. Do you
think it would have been better to make one big raid10? What if it was
raid5? And what if it was only 6 drives total?
Jim C. Nasby, Database Consultant decibel(at)decibel(dot)org
Give your computer some brain candy! www.distributed.net Team #1828
Windows: "Where do you want to go today?"
Linux: "Where do you want to go tomorrow?"
FreeBSD: "Are you guys coming, or what?"
In response to
pgsql-performance by date
|Next:||From: Josh Berkus||Date: 2004-09-16 21:05:28|
|Subject: Re: Question about PG on OSX|
|Previous:||From: Qing Zhao||Date: 2004-09-16 20:43:58|
|Subject: Re: Article about PostgreSQL and RAID in Brazil|