On Thu, Feb 23, 2012 at 1:07 PM, Alessandro Gagliardi
>> ...Apparently the last four columns don't exist in my database. As for the
>> first four, that is somewhat illuminating....
>> Then you are not running a current version of PostgreSQL so the first step
>> to performance enhancement is to upgrade. (As a general rule - there are
>> occasionally specific cases where performance decreases.)
> We're using 9.0.6. Peter, how do you feel about upgrading? :)
9.1's in beta; we're working on writing an upgrade system before
calling it GA, but it works fine. Feel free.
My hunch is still that your issue is lock contention.
> No, not average. I want to be able to do 100-200 INSERTs per second (90% of
> those would go to one of two tables, the other 10% would go to any of a
> couple dozen tables). If 1% of my INSERTs take 100 ms, then the other 99%
> must take no more than 9 ms to complete.
> ...actually, it occurs to me that since I'm now committing batches of 1000,
> a 100ms latency per commit wouldn't be bad at all! I'll have to look into
> that.... (Either that or my batching isn't working like I thought it was.)
We have many customers who do much more than this throughput, though
I'm not sure what level of resourcing you're current at. You might
consider experimenting with a larger system if you're having
In response to
pgsql-performance by date
|Next:||From: Steve Crawford||Date: 2012-02-23 21:37:54|
|Subject: Re: set autovacuum=off|
|Previous:||From: Jayashankar K B||Date: 2012-02-23 21:09:10|
|Subject: Re: [PERFORM] Disable-spinlocks while compiling postgres 9.1 for
ARM Cortex A8 |