Re: amazon ec2 - Mailing list pgsql-performance
| From | Greg Spiegelberg |
|---|---|
| Subject | Re: amazon ec2 |
| Date | |
| Msg-id | BANLkTi=nTpG=bS+rAONQi_ZDorBMwUzN1Q@mail.gmail.com Whole thread Raw |
| In response to | Re: amazon ec2 (Alan Hodgson <ahodgson@simkin.ca>) |
| Responses |
Re: amazon ec2
|
| List | pgsql-performance |
On Tue, May 3, 2011 at 2:09 PM, Alan Hodgson <ahodgson@simkin.ca> wrote:
On May 3, 2011 12:43:13 pm you wrote:Trying to, yes.
> On May 3, 2011, at 8:41 PM, Alan Hodgson wrote:
> > I am also interested in tips for this. EBS seems to suck pretty bad.
>
> Alan, can you elaborate? Are you using PG on top of EBS?
>
Let's see ...
EBS volumes seem to vary in speed. Some are relatively fast. Some are really
slow. Some fast ones become slow randomly. Some are fast attached to one
instance, but really slow attached to another.
I ran pgbench tests late last year comparing EC2, GoGrid, a 5 year-old lab server and a new server. Whether I used a stock postgresql.conf or tweaked, the current 8.4 or 9.0, or varied the EC2 instance size EC2 was always at the bottom ranging from 409.834 to 693.100 tps. GoGrid's pgbench TPS numbers in similar tests were, on average, 3X that of EC2 (1,399.550 to 1,631.887 tps). The tests I conducted were small with 10 connections and total 5,000 transactions. The single variable that helped pgbench tests in EC2 was to select an instance size where the number of cores was equal to or greater than the number of connections I used in the tests however this only improved things slightly (715.931 tps).
For comparisons purposes, I ran the same tests on a 24-way X5650 with 12 GB and SAS RAID 10. This server typically ranged from 2,188.348 to 2,216.377 tps.
I attributed GoGrids superior performance over EC2 as EC2 simply being over-allocated but that's just speculation on my part. To test my theory, I had wanted to put the database on a ramdisk, or like device, in EC2 and GoGrid but never got around to it.
Fast being a relative term, though. The fast ones seem to be able to do maybe
400 random IOPS. And of course you can only get about 80MB/sec sequential
access to them on a good day.
Which is why I'm interested in how other people are doing it. So far EC2
doesn't seem well suited to running databases at all.
I was doing this perhaps to convince management to give me some time to validate our software (PG backed) on some of the cloud providers but with those abysmal numbers I didn't even bother at the time. I may revisit at some point b/c I know Amazon at least has been making architecture adjustments and updates.
Greg
pgsql-performance by date: