I would also add that AWS' I/O capabilities are quite poor and expensive. I assume that you have tried purchasing additional IOOPs on that setup to see whether you got an expected speed up? If not you should try that as a diagnostic tool even if you wouldn't want to pay that on an ongoing basis.
We have a client that is I/O write bound and it has taken us significant efforts to get it to perform well on AWS. We definitely run our own instances rather than depend on RDS and have always been able to outperform RDS instances which seem to really be focused to provide a PAAS capability for developers who really don't want to have to understand how a db works. Running our identical environment on bare metal is like night & day under any circumstances when compared to AWS.
Client's requirement is AWS so we keep working on it and we like AWS for many things but understand it will always underperform on I/O.
Post actual measurements with and without IOOPs or create your own PG server instance and then people might be able to give you additional insights.
- - Ben Scherrey
On 2018-06-18 18:43:06 -0300, Juan Manuel Cuello wrote:
> I'm experiencing high WriteLatency levels in a Postgres server 9.3.20
> hosted in Amazon RDS.
A lot of performance improvements have been made since 9.3, and it'll
soon-ish be out of support.
If you can reproduce the issue on postgres proper, rather than a
modified version in an environment that precludes getting detailed data,
we might be able to sensibly help you further.
> So far it's been almost two months of investigation
> and people at AWS technical support don't seem to find the cause. I think
> it could be related to Postgres and the number of schema/tables in the
> database, that's why I post this issue here.
There've been improvements made since 9.3. Upgrade.
Greetings,
Andres Freund
