On 13 Jun 2013, at 11:43, Radim Vansa <rvansa(a)redhat.com> wrote:
After Pedro's question I have tried to run ISPN-3221 fix with
both old an new bundler (in configuration), and the result pretty surprises me as there
should not be much of a difference (as all messages that are part of the test should be
OOB => DONT_BUNDLE).
[1]
113842c8cf91cbb5a1bbd26e05fab7024fdec081 (last OK build)
new bundler: 1.5M reads/s, 7800 writes/s
old bundler: 1.3M reads/s, 4300 writes/s
So the new bundler over-performed the old one even before pedro's fixed? For your
first email I understood the opposite.
[2]
73da108cdcf9db4f3edbcd6dbda6938d6e45d148 (removing DONT_BUNDLE)
new bundler: 1.2M reads/s, 4800 writes/s
old bundler: 150k reads/s, 250 writes/s
This makes sense as the old bundler waits for 30 millis on every sync call. Also makes me
think that in [1] the results are swapped, i.e. 7800 writes/sec are with the *old*
bundler
[3]
ISPN-3221 fix (Pedro's branch)
new bundler: 1.5M reads/s, 6050 writes/s
old bundler: 1.3M reads/s, 3100 writes/s
[4]
ISPN-3221 cherry-picked onto 73da108cdcf9db4f3edbcd6dbda6938d6e45d148
new bundler: 1.5M reads/s, 7800 writes/s
old bundler: 1.3M reads/s, 4300 writes/s
This is practically the 5.2 code, at least
when it comes to bundling.
If the report is correct then if we apply pedro's patch with the new bundler we get
[3]/1:
- new bundler: 1.5M reads/s, 6050 writes/s
and the last ok build [1]/2 having the old bundler (i.e. 5.2.x config) had:
- old bundler: 1.3M reads/s, 4300 writes/s
In other words there's no performance regression, but the contrary.
BTW what's the performance of 5.2.x branch with which we should compare against.
Cheers,
--
Mircea Markus
Infinispan lead (
www.infinispan.org)