Peter Geoghegan <peter@2ndquadrant.com> writes:
> I'm not quite comfortable recommending a switch to milliseconds if
> that implies a loss of sub-millisecond granularity. I know that
> someone is going to point out that in some particularly benchmark,
> they can get another relatively modest increase in throughput (perhaps
> 2%-3%) by splitting the difference between two adjoining millisecond
> integer values. In that scenario, I'd be tempted to point out that
> that increase is quite unlikely to carry over to real-world benefits,
> because the setting is then right on the cusp of where increasing
> commit_delay stops helping throughput and starts hurting it. The
> improvement is likely to get lost in the noise in the context of a
> real-world application, where for example the actually cost of an
> fsync is more variable. I'm just not sure that that's the right
> attitude.
To me it's more about future-proofing. commit_delay is the only
time-interval setting we've got where reasonable values today are in the
single-digit-millisecond range. So it seems to me not hard to infer
that in a few years sub-millisecond values will be important, whether or
not there's any real argument for them today.
regards, tom lane
