Yeah, the reduced syscall change is in. He says the numbers are the
same between the two protocols as of that change.
I don't think it'd be very hard to contrive a test that would at least
give us a feel as to whether it's faster or slower. Pretty much
anything that could run for five minutes and do the ``same thing'' would
work if we just looked at the user time it used. Probably not
considerably less effort than using a profiler, though.
Was thinking of exactly that. Run the same number of gets/sets through
text or binary and compare the cpu usage reported via 'stats'. Should
also log latency differences in the clients. Between latency and CPU
used for batch operations, that'll give us most of the idea of if it
sucks or not.
Nah, but I trust my tests to report when something goes wrong, so I'd
like to keep it going. I don't mind fixing stuff, either.
Sweet :)
-Dormando
