Craig Ringer wrote:
I agreed with almost your entire post, but there is one sentence
with which I take issue.
> However, it will also increase latency for service for those
> workers because they may have to wait a while before their
> transaction runs, even though their transaction will complete much
> faster.
My benchmarks have shown that latency also improves. See these
posts for my reasoning on why that is:
http://archives.postgresql.org/pgsql-performance/2009-03/msg00138.php
http://archives.postgresql.org/pgsql-performance/2010-01/msg00107.php
So even though there is greater latency from the attempt to *start*
the transaction until it is underway, the total latency from the
attempt to start the transaction until *completion* is less on
average, in spite of the time in queue. Perhaps that's what you
were getting at, but it sounded to me like you're saying you
sacrifice latency to achieve the throughput, and that isn't what
I've seen.
-Kevin