[Pgpool-general] pgpool-II load balance performance

[Mikhail Kolesnik]

Hello.

I've been playing with pgbench and custom queries on pgpool-II-2.0.1 for
some time.

A few nodes running postgresql-8.2.6 are on the same LAN and
connected with Gigabit Ethernet. pgpool tself is one one node and two
other nodes are running Slony-I. I've also tried custom read-only
requests with pgbench, so Slony-I is not replicating at the time
queries are executed.

From my tests performance of pgpool in master/slave mode with 2 nodes is
near 85% of a _single_ node performance on the 'tpc-b'-like queries,
simple 'select' or complex 'select' with heavy sorting and few
additional conditions. 

But it performs quite well (almost 2 times faster on two nodes) on the
index-related operation like 'SELECT count(1) FROM ...'.

Sample db has ~500K records ('pgbench -i -s 5'). Changing number of
running concurrent clients and requests per client does not change the
whole picture.

I'd like to know what kind of requests are known to be reasonably well
balanced, or what are the actual performance ratios(compared to single
host) are you experiencing with existing pgpool
installations/applications?

Can also someone explain what 'connection_cache' in config file stands
for, as I can't understand that from the documentation?

-- 
Mikhail Kolesnik
ICQ: 260259143
IRC: mike_k at freenode/#crux, rusnet/#yalta