Re: postgres performance tunning

From: Scott Marlowe <scott(dot)marlowe(at)gmail(dot)com>
To: Greg Smith <greg(at)2ndquadrant(dot)com>
Cc: selvi88 <selvi(dot)dct(at)gmail(dot)com>, pgsql-performance(at)postgresql(dot)org
Subject: Re: postgres performance tunning
Date: 2010-12-20 18:19:59
Message-ID: AANLkTi=4P0boQriciZDE=vzjzMeLXVsO3C7xXShcbiT4@mail.gmail.com
Views: Raw Message | Whole Thread | Download mbox | Resend email
Thread:
Lists: pgsql-performance

On Mon, Dec 20, 2010 at 10:49 AM, Greg Smith <greg(at)2ndquadrant(dot)com> wrote:
> Scott Marlowe wrote:
>>
>> I can sustain about 5,000 transactions per second on a machine with 8
>> cores (2 years old) and 14 15k seagate hard drives.
>>
>
> Right.  You can hit 2 to 3000/second with a relatively inexpensive system,
> so long as you have a battery-backed RAID controller and a few hard drives.
>  Doing 5K writes/second is going to take a giant pile of hard drive or SSDs
> to pull off.  There is no possible way to meet the performance objectives
> here without a lot more cores in the server and some pretty beefy storage
> too.

And it gets expensive fast as you need more and more tps capability.
Those machines listed up there were $10k two years ago. Their
replacements are $25k machines with 48 cores, 128G RAM and 34 15k hard
drives, and they get about 8k tps. Note that due to the nature of
these machines' jobs they are NOT tuned heavily towards tps in real
life, but more for handling a bunch of little reads and few big writes
and reads simultaneously. The new machines are much more than 30 or
40% faster in real world testing, for our workload they're about 10x
as fast, since we were CPU bound before with 8 cores.

In response to

Browse pgsql-performance by date

  From Date Subject
Next Message Robert Haas 2010-12-20 18:57:09 Re: UNION and bad performance
Previous Message Jeremy Harris 2010-12-20 17:59:42 Re: CPU bound