From: Fei Liu <fei.liu@aepnetworks.com>
To: Scott Marlowe <smarlowe@g2switchworks.com>
Cc: Merlin Moncure <mmoncure@gmail.com>
Cc: Andreas Haumer <andreas@xss.co.at>
Cc: pgsql-performance@postgresql.org
Subject: Re: Query performance problems with partitioned tables
Date: Tue, 08 May 2007 14:41:38 -0400
Message-ID: <4640C462.2040909@aepnetworks.com> (raw)
In-Reply-To: <1178291472.2953.47.camel@state.g2switchworks.com>
References: <4635D678.3050205@xss.co.at>
<4639F97D.3080003@aepnetworks.com>
<b42b73150705031937g7884ae55ha9cc670e26a27340@mail.gmail.com>
<1178291472.2953.47.camel@state.g2switchworks.com>
Scott Marlowe wrote:
> On Thu, 2007-05-03 at 21:37, Merlin Moncure wrote:
>
>> On 5/3/07, Fei Liu <fei.liu@aepnetworks.com> wrote:
>>
>>> Hello, Andreas, I too am having exactly the same issue as you do.
>>> Comparing my partitioned and plain table performance, I've found that
>>> the plain tables perform about 25% faster than partitioned table. Using
>>> 'explain select ...', I see that constraints are being used so in
>>> partitioned tables fewer rows are examined. But still partitioned tables
>>> are 25% slower, what a let down.
>>>
>> That's a little bit harsh. The main use of partitioning is not to
>> make the table faster but to make the maintenance easier. When
>> constraint exclusion works well for a particular query you can get a
>> small boost but many queries will break down in a really negative way.
>> So, you are sacrificing flexibility for easier maintenance. You have
>> to really be careful how you use it.
>>
>> The best case for partitioning is when you can logically divide up
>> your data so that you really only have to deal with one sliver of it
>> at a time...for joins and such. If the OP could force the constraint
>> exclusion (maybe by hashing the timestamp down to a period and using
>> that for where clause), his query would be fine. The problem is it's
>> not always easy to do that.
>>
>
> Agree++
>
> I've been testing partitioning for a zip code lookup thing that was
> posted here earlier, and I partitioned a 10,000,000 row set into about
> 400 partitions. I found that selecting a range of areas defined by x/y
> coordinates was faster without any indexes. The same selection with one
> big table and one big (x,y) index took 3 to 10 seconds typically, same
> select against the partitions with no indexes took 0.2 to 0.5 seconds.
>
> For that particular application, the only way to scale it was with
> partitioning.
>
In my particular case, I have 2 million records uniformly split up in 40
partitions. It's ranged data varying with time, each partition has one
month of data. Do you think this is a good candidate to seek performance
boost with partitioned tables?
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: fei.liu@aepnetworks.com, smarlowe@g2switchworks.com, mmoncure@gmail.com, andreas@xss.co.at
Subject: Re: Query performance problems with partitioned tables
In-Reply-To: <4640C462.2040909@aepnetworks.com>
* 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