Discussion:
[PATCH 0 of 3] Begin benchmarks with "burn-in"
Martin Geisler
2008-12-12 00:42:29 UTC
Permalink
Hi,

In trying to explain some measurements, Jesper felt paranoid and asked
if we could start the benchmarks with a warm-up phase. The idea is to
"burn-in" the CPU and network so that everything is to tuned when the
benchmark begins for real.

These patches add this, but I think there is something wrong with the
network burn-in. I tried to make it measure the round trip time by
letting everybody send ping packets back and forth.

With three players I get numbers like this:

RTT 1 <-> 2: min avg max = 3447.198 ms 3899.702 ms 4363.692 ms
RTT 1 <-> 3: min avg max = 2529.485 ms 3307.890 ms 3892.167 ms
RTT 1 <-> *: min avg max = 2529.485 ms 3603.796 ms 4363.692 ms

This should be the round trip time *per* packet when sending 10,000
packets. But this is clearly too large -- and if I send fewer packets
the average time drops. So I guess I've programmed it wrong, somehow.

Please take a look at this code, experiment with it and let me know
what you think. Thanks.

Loading...