Dimitri <dimitrik(dot)fr(at)gmail(dot)com> wrote:
> Of course the Max throughput is reached on the number of users equal
> to 2 * number of cores
I'd expect that when disk I/O is not a significant limiting factor,
but I've seen a "sweet spot" of (2 * cores) + (effective spindle
count) for loads involving a lot of random I/O.
> So, do I really need a pooler to keep 256 users working??
I have seen throughput fall above a certain point when I don't use a
connection pooler. With a connection pooler which queues requests
when all connections are busy, you will see no throughput degradation
as users of the pool are added. Our connection pool is in our
framework, so I don't know whether pgbouncer queues requests.
(Perhaps someone else can comment on that, and make another suggestion
if it doesn't.)
In response to
pgsql-performance by date
|Next:||From: Joshua D. Drake||Date: 2009-05-12 16:07:59|
|Subject: Re: Any better plan for this query?..|
|Previous:||From: Dimitri||Date: 2009-05-12 15:41:14|
|Subject: Re: What is the most optimal config parameters to keep stable write TPS ?..|