Stephen,

A frequent culprit in that sort of performance is the garbage
collect thread, and the clues can be found in a gstat -a -r. Look for
tables with large numbers of back versions and indexes with > 32,000
instances of a single value. That combination puts the garbage collect
thread into a state where it doesn't get enough time in any one quantum
to do anything valuable. The resolution is to drop those indexes and
replace them with more selective indexes - adding the primary key at the
end of the list of fields to be indexed for example.

Regards,


Ann