Re: WIP patch for latestCompletedXid method of computing snapshot xmax

Josh Berkus <josh(at)agliodbs(dot)com> writes:
> Hmmm. Your results are withing the margin of error for DBT2, so they
> show no real difference. What we need for this is a heavy-read
> workload, though; on a workload like DBT2 (which is mostly writes) I
> wouldn't expect lazy-XID to help much.

Lazy-XID doesn't help on a write-mostly workload, but I would have
expected to see some benefit from the latestCompletedXid patch.

The rough tests I just finished suggest that the win to look for is
improvement of the very tail of the distribution --- well beyond the
90th percentile point which is the most we can see in Markus'
printout. Can we get 95% and 99% response time percentiles?

For comparison, in my little test I got 0.828546 vs 0.873957 as the
99% percentile pgbench transaction times.

regards, tom lane