Solaris x86 client? [No]

[Matt23]

I check back on Folding@Home about every year or so to see if there's a Solaris x86 client. With a Windows client, and a Linux client, it's kind of a head-scratcher that there's no Solaris x86 client. Solaris' code is open now, and it runs on quite a few x86 hardware platforms. And now there's a Playstation client?! What's more, OpenSolaris continues to grow in popularity. And unlike Linux, there aren't 117 distributions each with wacky kernel and linking gotchas.

Now I know people can emulate Windows or Linux to some degree on Solaris x86 but why should we have to? Solaris x86 is open now, and you can use GCC or the Sun Compiler (also free) to build solid code. It's a waste of CPU cycles (and in effect, electricity) to emulate another OS for something like this.

The SETI search project is interesting but I think most of us who are 'fans' of the Folding@home project see F@h as a bit more down to earth, pun not necessarily intended. The SETI folks have clients for almost any platform you can think of. That's one of the reasons it's so successful and widespread. Imagine if only a fraction of that CPU time could be dedicated to something like F@h.

Heck I hate to say it, but has anyone even considered a Java (!) client? Java sucks a lot less than it used to, and RAM is relatively cheap these days.

Anyway, if someone needs some time on a Solaris x86 box and would be willing to do some development, I can provide shell access to one, etc etc.

thanks for your time!
-Matt

[bruce]

Thanks for your patience, but it's probably misplaced. The Pande Group is concentrating on extremely high performance clients using parallel techniques for hardware such as SMP and GPU and Cell and although the standard client is still an important part of the choices, it's not too likely that they'll adapt it to another platform.

I proably was the one that answered your question last year, and nothing has really changed in relation to Solaris. I suspect that the only hope is that you find a WIndows or Linux emulator that runs on Solaris. The fundamental code for x86 would run quite well in either type of emulator because of the preponderance of arithmetic code.

Support of any new platform requires sufficient in-the-lab hardware to test each new version plus extra time to debug the features that are modified by the porting -- and that would include several clients and a multitude of cores. It can be an expenxive support job for a relatively few added CPUs. (From a sheer numbers point of view, the existing clients already cover the OSs found on a very high percentage of home computers.)

[RAH]

And it could work backward too, to some extent. Much of the folding software is open source.
I believe I read that you can get a open client, that should run on your own compilation of gromacs, etc.

Why don't you check it out, and see if you can get it to compile. If you do, maybe Stanford would look
at making a official client. I sort of wish that I knew something about software writing. It would
be fun to see if you could get something to work.

With all the fires that they have right now, I don't think it will happen from here.

[sneakers55]

The SETI folks have clients for almost any platform you can think of.

There are SETI clients for almost any platform because they support ports for almost anything that will talk to the Internet.

SETI doesn't trust their client ports much because they run everything three times and vote the odd man off the island.

That's one of the reasons it's so successful and widespread.

SETI "successful"? They run about 387 teraflops (divide by three to get something comparable) while F@H regularly breaks a petaflop.

Imagine if only a fraction of that CPU time could be dedicated to something like F@h.

The great majority of that CPU time is most likely Windows, Mac, and Linux. All of which run under F@H.

[7im]

Rattledagger should be along shortly to post the stats for Solaris as a percentage of total processing power for SETI.

I don't recall the exact number, and RD will correct me, but x86 hardware with Windows and Linux is around 90%, and solaris of any type is ~1%.

Stanford has limited developmental resources, so they have to get the most science bang for their development buck. They are picking the low hanging fruit, with high performance hardware (SMP, GPU, PS3) and with the largest participant groups (x86 on Windows, Linux, etc.)

Sorry. Unless Solaris sees a resurgence in the tech market, there may never be a Solaris client.

[alancabler]

If SPARC machines could produce 1% of project power, that'd be a lot of TFLOPS, but still only 1% change.

Unless Solaris sees a resurgence in the tech market, there may never be a Solaris client.

Apple/IBM bears witness to the truth of that statement... times change, technology falls out of favor and sometimes disappears.
@ Matt23,
Could you try a week or so of single- core emulation and post your results?
Demonstrated folding horse- power might make a future multi- core SPARC client more compelling.
Results trump rants.

[7im]

The new GPU2 client just increased TFLOP output by a little over 7%, so by comparison, I guess that would be time well spent. When the GPU2 and SMP clients stablize (release to production) maybe there will be time to revisit this issue. Probably not until then.

[v00d00]

Virtualbox now appears to have solaris builds, so maybe you can get something running via that. It will probably have more now that Sun owns it.

[DavidYuen]

F@H client could run in Opensolaris 0906 under Wine. Download wine 1.0 (yes 1.0.x) from Opensolaris package website and Folding@home-Wine32-x86.exe from F@H website
Then
(1) install wine 1.0 package
(2) run
$ wine Folding@home-Wine32-x86.exe.

It worked in my rig for most versions of wine (including SFEwine)