SPARC Solaris Einstein@Home?

Anonymous

22 Aug 2005 15:32:03 UTC

Topic 13088

(moderation:

)

We had a voulunteer for doing the port, but I haven't heard of him for months now. Apparently he ran out of free time for that (like we all do...). Currently there is done quite some work on the code of BOINC as well as on the science code. Maybe we'll start another attempt when the codebase is more stable again. But the priority for new ports is rather low for now.

Bernd Machenschalk

Joined: 15 Oct 04

Posts: 2,684

Credit: 25,950,161

RAC: 34,820

SPARC Solaris Einstein@Home?

10 Jan 2006 5:00:37 UTC

Message 2680

(moderation:

)

I just built a Solaris/SPARC version of the new Albert App, which seems to run well on Solaris 7 and 10 (and thus it should on 8 and 9, too). We plan to release it after some testing. Our SPARCs are somewhat slow, so this may take a while. Stay tuned.

Bruce Allen

Joined: 15 Oct 04

Posts: 958

Credit: 170,849,008

RAC: 0

The Solaris Einstein@Home app

14 Jan 2006 17:50:46 UTC

Message 2681

(moderation:

)

The Solaris Einstein@Home app that Bernd built is now being distributed (command line version only; graphical version coming soon). Please use this thread to report success or problems.

Bruce

Bernd Machenschalk

Joined: 15 Oct 04

Posts: 2,684

Credit: 25,950,161

RAC: 34,820

The stock clients all request

15 Jan 2006 9:11:48 UTC

Message 2682

(moderation:

)

The stock clients all request sparc-sun-solaris2.7 platform, as they are built on a Sol7 so they should run on all systems from then on. If you compile your own client, you are advised to use --build=sparc-sun-solaris2.7 as additional configure option, so the client will report the same platform.

Due to an old build process (which has been improved since then) the old 4.19 links to shared libraries that are not present on all systems (to say the least), in particular libstdc++.so.3 and libgcc_s.so.1. If you have a gcc installed, you can create a symlink named libstdc++.so.3 to your version of libstdc++.so.

I am running the stock recommended 4.43 client without problems.

Bernd Machenschalk

Joined: 15 Oct 04

Posts: 2,684

Credit: 25,950,161

RAC: 34,820

RE: Hi... I'm very happy to

15 Jan 2006 15:58:48 UTC

Message 2683

(moderation:

)

Quote:

Hi...
I'm very happy to have one application for my solaris (other than seti)...
But at this time i have problems whit the computing... some WUs come with no time (and they stay with 0% in my BViewer...)...
is that a WU problem or a Application problem ?

Which client are you using? I don't know about BViewer, but what do you get when you grep client_state.xml for "fraction_done" on the machine in question?

Bernd Machenschalk

Joined: 15 Oct 04

Posts: 2,684

Credit: 25,950,161

RAC: 34,820

RE: RE: But the

18 Jan 2006 13:29:20 UTC

Message 2684

(moderation:

)

Quote:

Quote:
But the performance looks rather poor:

As it is visible on the above result page, the processing is clearly to slow compared to usual PC systems.
Not two complete CPU days, but comes near --- do you took any optimization measures at all? What compiler do you used to build the client? gcc 3.0, 3.3, 3.4, 4.0, Sun Studio 8,9,10,11? There are different possibilities to get the client faster, on SPARC-Solaris sometimes the program may run even faster with -O2 than -O3 with gcc...

I am aware of the rather poor performance and am working on optimizations. I was happy to get something working at all. The App has been compiled with gcc-3.3 on a Solaris7 system, for now with -O3. I'm investigating and expect to improve the performance significantly.

Bernd Machenschalk

Joined: 15 Oct 04

Posts: 2,684

Credit: 25,950,161

RAC: 34,820

RE: RE: I am aware of

18 Jan 2006 14:11:35 UTC

Message 2685

(moderation:

)

Quote:

Quote:

I am aware of the rather poor performance and am working on optimizations. I was happy to get something working at all. The App has been compiled with gcc-3.3 on a Solaris7 system, for now with -O3. I'm investigating and expect to improve the performance significantly.

BM

Okay, than for me all is clear now. I propose usage of gcc 4.0.2 with -ftree-vectorize and maybe the VIS extension (which would limit usage to UltraSPARC systems and higher, what makes sense in matters of acceptable performance) -mvis.

AFAIK the -ftree-vectorize doesn't make sense without the extensions (i.e. is ignored). Also it doesn't help much with our code, at least the 4.0.1 on x86 didn't recognize our loops as being vectorizable.

Quote:

I can't recommend usage of the Sun Studio compiler (though 11 is free to be used), because that would require BOINC too to be compiled with it; gcc libs can't be used in Sun compiler builds. The last time I tried BOINC to compile with a Sun compiler it looked rather difficult to get through.

Hasn't gotten easier yet...

Bernd Machenschalk

Joined: 15 Oct 04

Posts: 2,684

Credit: 25,950,161

RAC: 34,820

RE: AFAIK the

18 Jan 2006 16:13:15 UTC

Message 2686 in response to message 2685

(moderation:

)

Quote:

AFAIK the -ftree-vectorize doesn't make sense without the extensions (i.e. is ignored). Also it doesn't help much with our code, at least the 4.0.1 on x86 didn't recognize our loops as being vectorizable.

Sorry, must have been gcc 4.0.2.

Bernd Machenschalk

Joined: 15 Oct 04

Posts: 2,684

Credit: 25,950,161

RAC: 34,820

RE: bit now with 2 projects

18 Jan 2006 23:46:57 UTC

Message 2687

(moderation:

)

Quote:

bit now with 2 projects and two processors, the boinc-client seams to have a problem....
any idea ?

Hm, I'm tempted to suggest a newer client, but I'm aware there is no such available from BOINC for download - you'll either have to roll your own or look for other friendly crunchers who did that for you...

What happens if you set your preferences to "use no more than 1 CPU"?

Bernd Machenschalk

Joined: 15 Oct 04

Posts: 2,684

Credit: 25,950,161

RAC: 34,820

RE: I got another problem

19 Jan 2006 14:37:32 UTC

Message 2688

(moderation:

)

Quote:

I got another problem with the new app.
I'm using boinc 4.42 (I know I should upgrade, but I'm too lasy and had no problems till now.)

I'm running both SETI and albert on the Solaris boxes.
Problem here is, the app does not really stop, if boinc asks the app to do so.
as well I tried to kill the app normal and it was still running.
kill -9 worked.

Any idea?
Thanks

Achim

What version of Solaris are you running this on? Are the patches up-to-date?

Anybody else seen this?

It might be a problem with the Core Client, or a with shared library (pthreads?).

Bernd Machenschalk

Joined: 15 Oct 04

Posts: 2,684

Credit: 25,950,161

RAC: 34,820

RE: SPARC optimizations

15 Feb 2006 12:15:01 UTC

Message 2689

(moderation:

)

Quote:

SPARC optimizations with GCC. I think this may help future development concerning Sparcs if it wasn't done already.

Article from OSNews.

SPARC Optimizations

Thanks, this looks useful. However I just checked again - -mcpu=ultrasparc doesn't gain anything with the current code. I expect some speedup once I get the autovect feature of GCC 4 to recognize our loops as vectorizable and use vis, but this may take some time.

SPARC Solaris Einstein@Home?

Forums › Wish List

Comment viewing options

Forums › Wish List