Performance and low CPU usage

[dave irving]

Hi,

Im currently evaluating mysql cluster - and trying to see the maximum performance I can get out of it with various hardware set-ups.

Currently I've got a simple set-up: 2 data nodes, 2 mysqld nodes, 1 management node.
All the boxes are 2-CPU (each being dual core), running linux EL, and have gigabit network cards.

The queries I issue to the cluster are very simple (insert, delete by primary key, select by primary key).
I have several clients connecting in over JDBC.

Im noticing that performance seems to top out when there are around 60 (client) connections in to the cluster (irrespective of how many boxes these connections originate from).

The interesting thing is that nothing (at first glance) seems to be hurting: The mysqld nodes never get above 50% (total) CPU usage, the data nodes hardly sweat the CPU at all, and using ntop I can see that the network seems to have loads of free bandwidth too.

What should I be looking for next to find the bottle neck? (E.g, could it be transaction log configuration etc?).

Another interesting thing is that I've tried altering the two params which seem to crop up a lot in performance discussions (ndb_force_send=0, ndb_use_exact_count=0) - but these seem to -- reduce -- the overal performance I get with my tests.

Any advice of where to look next would be appreciated.

Cheers,

Dave