harmony-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Aleksey Shipilev" <aleksey.shipi...@gmail.com>
Subject Re: [performance] quick sort is 4x slower on Harmony
Date Wed, 16 Jan 2008 14:36:19 GMT
Hi, Egor, Pavel!

First, the impact of lock-less solution to other benchmarks. I had
measured two modes: (a) locks eliminated completely, (b) addNewValue
is locked only on insertion of new value into the table.

a. SPECjbb2005: no boost/degradation for overall score, (a) +10% score
improvement for 1st warehouse, (b) +9% score improvement for 1st
warehouse.

b. Dacapo: no boost/degradation for overall score on each of
subbenchmarks; (a) +9% improvement on first iteration, (b) + 5%
improvement on first iteration.

c. SciMark: no boost/degradation.

So I propose to apply patch for (b) and introduce two options:
 a. VP locking enable/disable
 b. VP flagged locking (to check Egor's suggestion).

> cool investigation, Alexey! I love you :)
Ah, I should see that comes :D

> Alexey, I would be glad to prepare the patch by myself, but I can only
> start doing it at the end of the next week. So, if you feel like doing
> it, please, take the task, I would gladly review it and commit.
Sorry, I was overloaded with my university exams :) I filed the JIRA
[1] with (b) patch attached, but I can't imagine when I could get to
implementation of flagged collision check and externalization of the
options. Egor, can you elaborate with them?

Thanks,
Aleksey.

[1] https://issues.apache.org/jira/browse/HARMONY-5396

Mime
View raw message