harmony-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Pons Rudy <rudy.p...@etu.u-bordeaux1.fr>
Subject [GSoC] Prosposal for HARMONY-6445
Date Fri, 16 Apr 2010 21:31:27 GMT

I have sent my proposal for Dacapo benchmark suite investigation, but  
haven't sent it to the list to discuss before. After discussion with  
potential mentor, I send it.

I think I'll be able to correct bugs and improve performances of some  
parts of the API. I'm not sure either to be able to significantly  
improve all the parts of the API tested by Dacapo. Dacapo have several  
tests. It will be useful to know on which test you think I should  
focus. I will test execution time of each test and compare them with  
RI, and identify which part of the API are used by each benchmark  
test, but if you want to take a look before, each test is very briefly  
described here : http://www.dacapobench.org/benchmarks.html

Here's my proposal. If you have any comment...


Proposal Title: [Harmony][performance] Investigate dacapo benchmark suite

Student Name: Rudy Pons

Student E-mail: rudy.pons@etu.u-bordeaux1.fr

Student Major: Computer Science

Student Degree: Studying for Graduate of Science Degree, 2nd year

Organization/Project: Apache Software Fondation/Harmony Project

Assigned Mentor: Mark Hindess

Proposal Abstract:

The goal of the project is to fix bugs and improve Harmony API global  
performance by investigating results and performance of the Dacapo  
benchmark suite on Harmony.

Detailed Description:

Investigate the performance of the Dacapo benchmark suite  
(http://www.dacapobench.org/) on Harmony. Some of the benchmarks don't  
run and some are unstable, and some run slower than the RI.. This task  
would be to figure out why they don't run, what could be done to make  
them more stable and how to make them run faster.

First step will be to run Dacapo suite, find which test don't run or  
are unstable, read Dachapo source code to understand why they don't  
work fine. Exploring JIRA tickets related to modules used in these  
tests should be useful too. Then, I will be able to list all broken  
features involved in the Dacapo suite, begin to read appropriate  
Harmony code, produce patches to fix these problems, and add  
regression tests.

Second step will be to improve performance. For this, I will focus on  
tests where Harmony is slower than RI, try to write some new tests to  
understand what precise parts of Harmony are slower than RI for this  
test. I will sort all of these part to find the slowest, and become  
familiar with involved Harmony code, to find how many time should take  
each part to obtain significant performance improvement. Then I will  
focus on best parts to work on, looking possible performance  
improvement for code time ratio, and use of the API. Then I will try   
different modifications on Harmony, and compare the results between  
these modifications on Dacapo benchmark.

I?ve read part of the Dacapo source code, and begin to fix bugs on the  
H2 test.

Available Timeline:

I have exams to attend and will not be fully available for the  
Community Bonding Period before the end of these exams, May 11th.
I have no other unavailability during GSoC.

Draft Timeline:

Week 1: Run Dacapo benchmark suite, read source code, and identify bugs.

Week 2-5: Explore Harmony source code, write patch to fix identified  
bugs and add regression tests.

Week 6: Write shorter tests than Dacapo ones to identify precisely  
slowest part in Harmony.

Week 7-11: Write and compare several optimisations for different parts  
of the API.

(This timeline is only a draft, as I can't predict how many bugs will  
be discovered, and so how many time will be necessary to fix them)

Additional Information:

I?ve been using Java since 3 years, and have worked on several  
projects based on Java, including a file conversion utility, a tennis  
tournament draws manager, a cheap online game, and some AI for games.

I?m 19 years old, living in Bordeaux, France, and studying in Bordeaux  
1 University.



Rudy Pons

View raw message