> On Sat, Oct 5, 2013 at 6:10 PM, Noah Misch <noah@leadboat.com> wrote:
>> The sum of the squares of the latencies wraps after 2^63/(10^12 * avg_latency
>> * nclients) seconds. That's unlikely to come up with the ordinary pgbench
>> script, but one can reach it in a few hours when benchmarking a command that
>> runs for many seconds. If we care, we can track the figure as a double. I
>> merely added a comment about it.
>
> Using a double seems wise to me.
I think that both int64 & double are fine.
The likelyhood of having underflows (double) or overflows/wraparounds
(int64) seems pretty low for any practical run. I took the former because
it is exact... up to when it is totally wrong. The later one may provide
underestimated results silently: it would fail more continuously.
So I'm okay if it is changed to double consistently.
--
Fabien.
