Gentoo Websites Logo
Go to: Gentoo Home Documentation Forums Lists Bugs Planet Store Wiki Get Gentoo!
Bug 126436 - Problem with optimizing sci-misc/boinc
Summary: Problem with optimizing sci-misc/boinc
Status: RESOLVED UPSTREAM
Alias: None
Product: Gentoo Linux
Classification: Unclassified
Component: Current packages (show other bugs)
Hardware: All Other
: High trivial (vote)
Assignee: Marcus D. Hanwell (RETIRED)
URL:
Whiteboard:
Keywords:
Depends on:
Blocks:
 
Reported: 2006-03-16 09:56 UTC by M. LaPlante
Modified: 2006-06-11 03:49 UTC (History)
2 users (show)

See Also:
Package list:
Runtime testing required: ---


Attachments

Note You need to log in before you can comment on or make changes to this bug.
Description M. LaPlante 2006-03-16 09:56:33 UTC
Currently, as far as I can tell, the Boinc client is built with full optimizations when emerged, as is other software (assuming it's specified in Make.conf).  Strangely, this may be a problem for Boinc.  The problem is that the project cores that run within boinc are not compiled on the local machine, and are therefore usually generic builds.  At the same time, it is the client that is responsible for benchmarking the machine.  The benchmark is used when determining the "credit" received by the client for completed work.  Therefore, if you run an optimized benchmark, but the generic cores, you will actually be penalized when credit is awarded by the various projects.  The benchmarks will reflect a faster machine than the core clients are able to take advantage of.
Comment 1 Jakub Moc (RETIRED) gentoo-dev 2006-03-16 10:29:13 UTC
I really fail to see why is this a Gentoo bug...
Comment 2 M. LaPlante 2006-03-16 14:43:58 UTC
Because it's specific to the Gentoo ebuilds...

Distros with precompiled apps obviously would not have this issue.
Comment 3 Jakub Moc (RETIRED) gentoo-dev 2006-03-17 13:12:56 UTC
(In reply to comment #2)
> Because it's specific to the Gentoo ebuilds...
> 
> Distros with precompiled apps obviously would not have this issue.

So, how upstream awards some credits it Gentoo's fault. Really makes sense. Well, so compile it for i386 with some stupid counter-optimizations and you'll get more credits.
Comment 4 Gerard Neil 2006-04-09 23:23:00 UTC
(In reply to comment #3)

> So, how upstream awards some credits it Gentoo's fault. Really makes sense.
> Well, so compile it for i386 with some stupid counter-optimizations and

Sure, the BOINC credit mechanism could be improved, but it's always going to be imperfect. That doesn't mean that the gentoo ebuild should make the problem worse!

The BOINC core client downloads and schedules other application binaries. The build environments of the core client and those other applications are all different, and the whole thing is cross platform, cross architecture. That's why assigning fair credits is a difficult problem. 

There's *no* advantage to building an optimised core client, because the program doesn't do the grunt work itself. Instead, an optimised client disadvantages gentoo users. On average, they'll get less credit for the same work. This will affect some projects more than others; it will be worst for a project like ClimatePrediction.net where the credit isn't averaged over redundant results. People like their credits. Why not be as fair as possible?

A fair solution would seem to be: use any optimisations that are supplied by upstream, but strip out the user-supplied ones. That would put gentoo boinc on a par with generic binaries.

By the way, this isn't just a gentoo problem. Any distribution which packages the core client as a binary faces the same issue: what optimisation to use when building it.
Comment 5 Jakub Moc (RETIRED) gentoo-dev 2006-06-11 03:49:18 UTC
Tell upstream that their crediting system sucks...