Of course, there are only a few PS3s running Linux.
FYI, I have been negotiating with Sony for 8 months
about porting BOINC to the standard PS3
(which runs a proprietary OS, not Linux)
and having it included in the default software,
like they're currently doing for Folding@home.

These negotiations stalled, at least for the time being.
Maybe some activity in the PS3/Linux/BOINC area will
rekindle their interest.

-- David

It'll be a very good thing if BOINC could run on PS3 OS like F@H. For most of people, it's to difficult to install Linux on PS3.

David please, try to relaunch negociations with Sony. I'm sure PS3 can bring a lot of power to all BOINC projects.

---

Package installed and launched.
All WUs directly under error. I stopped for the moment to not download WUs for nothing.
My PS3 computer page: PS3

You need to have libspe installed on your system. Download the rpms at:
http://www.bsc.es/plantillaH.php?cat_id=253

OK, but which version and how I can install it ?
Thanks.

Please download those rpms and install them using this command, as root: "rpm -i rpmfilename.rpm"

Continuing not working. Which file I have to take ?

---

Package installed and launched.
All WUs directly under error. I stopped for the moment to not download WUs for nothing.
My PS3 computer page: PS3

You need to have libspe installed on your system. Download the rpms at:
http://www.bsc.es/plantillaH.php?cat_id=253

OK, but which version and how I can install it ?
Thanks.

Please download those rpms and install them using this command, as root: "rpm -i rpmfilename.rpm"

Continuing not working. Which file I have to take ?

I believe you need to install ALL of those rpms. Please let me know, if that still doesn't work for you.

---

I believe you need to install ALL of those rpms. Please let me know, if that still doesn't work for you.

Yes !
My PS3 is now crunching its two first seti WU !

---

I succed to use client_gui of PS3GRID on Gaurav's boinc_client.
PS3GRID's boinc_client request work with an unrecognize system flag on S@H. So, it can't work except if S@H team accept to include this flag on the list.

---

I succeeded to use client_gui of PS3GRID on Gaurav's boinc_client.
PS3GRID's boinc_client request work with an unrecognized system flag on S@H. So, it can't work except if S@H team accept to include this flag on the list.

I wonder does this client have or need an app_info.xml file?
If It does It would need one, As this client and app looks like It would be a natural for the anonymous platform, Otherwise I'm not sure what to day.

---

The T1 Trust, PRR T1 Class 4-4-4-4 #5550, 1 of America's First HST's

My PS3 is now running S@H under PS3GRID BOINC distribution. You can read results here.
This BOINC version limit the number of CPU to one. So I run only one WU at a time. But I think it's probabely a good thing for the future when we'll optimize all parts of application for SPU.
For the moment, one task when only one is running, seems to be 1.5 times faster than one task when two are running.

---

is there a way to see ps3 chips stats on boincstats?

---

Performance seems to vary extremely on PS3

If i compare those --> PS3 WU true angle range is : 0.393932

http://setiathome.berkeley.edu/result.php?resultid=543766458

CPU time: 44375.916

and compare that one to --> EV67 @600MHz WU true angle range is : 0.393750

http://setiathome.berkeley.edu/result.php?resultid=538362714

CPU time: 41658.992031

Doesn't look good for the PS3 cuz most WUs have and AR of 0.420 or 0.393

Seems as if the PS3 code needs some optimizing :-)

---

Join BOINC United now!

Performance seems to vary extremely on PS3

If i compare those --> PS3 WU true angle range is : 0.393932

http://setiathome.berkeley.edu/result.php?resultid=543766458

CPU time: 44375.916

and compare that one to --> EV67 @600MHz WU true angle range is : 0.393750

http://setiathome.berkeley.edu/result.php?resultid=538362714

CPU time: 41658.992031

Doesn't look good for the PS3 cuz most WUs have and AR of 0.420 or 0.393

Seems as if the PS3 code needs some optimizing :-)

Sure we need to optimize. I can do it, but I block on seti compilation under Linux. Is someone, perhaps you, could give me the procedure with gcc, it'll be great.

Thanks,
ThierryH

---

Seems as if the PS3 code needs some optimizing :-)

Sure we need to optimize. I can do it, but I block on seti compilation under Linux. Is someone, perhaps you, could give me the procedure with gcc, it'll be great.

Thanks,
ThierryH

Those times indicate that the SPEs are not being used at all, so yes a huge amount of optimization remains to be done. I'd go back to first principles and build the Cell code from scratch using the linux code as a reference. I would even pull out the FFTW routines and rewrite them specifically for the SPEs. All the SPD should do is to just sit there and dish out transforms to whichever SPE is available. If one job is running, then six SPEs share the work. If two jobs are running, then they get an average of 3-3.5 SPEs to crunch for them.
With optimised SPE FFTs your 41000 secs should reduce to around 1000s! I'd expect a RAC of above 5000 for a PS3.
Gotta be worth someone working on! I'd do it myself if I had the time. :-(

Hi Andy,

the FFT part is actually not such a problem - FFTW has been ported to the PS3 (read: SPEs) already (even though it's alpha as of now).

Still, FFTs are not the only place a lot of time is spent. So your estimate of 1000s is a bit optimistic IMO, but here's hoping ;)

I've given Gaurav access to the development sources at lunatics.at - hope that will help.

BTW Gaurav, since this is a Linux build, you may want to use the 2.2B Linux sources instead of the 2.3 S5 Windows ones (since those need a bit of editing to get to compile on Linux).

Regards,
Simon.

---

Donate to SETI@Home via PayPal!

Optimized SETI@Home apps + Information

BTW, can someone here, state the typical range of FFT lengths that are used by the SETI app? Note that the Cell does well only for rather large values (see plots here). That is expected because of the Cell's design.

I'm using this same FFTW for my Cell binaries, with some modifications that improve performance for the mid-range FFT lengths.

---

Gaurav,

here's a link to a post by Joe Segur. He stated:

8, 16, 32, 64, 128, 256, 512, 1024, 2048, 4096, 8192, 16384, 32768, 65536, and 131072 lengths are all used.

Technically, some simple changes in the parameters sent in the workunit header could require additional lengths but it's highly unlikely the project will do that.

HTH,
Simon.

---

Donate to SETI@Home via PayPal!

Optimized SETI@Home apps + Information

Gaurav,

here's a link to a post by Joe Segur. He stated:

8, 16, 32, 64, 128, 256, 512, 1024, 2048, 4096, 8192, 16384, 32768, 65536, and 131072 lengths are all used.

Technically, some simple changes in the parameters sent in the workunit header could require additional lengths but it's highly unlikely the project will do that.

HTH,
Simon.

This explains things a lot, I think. The FFT based performance of the Cell begins to shine for only the last two values in this list. This may change to an extent, with FFTW improving in the near future, but I don't know by how much.

And then there is the question of improving the performance of the other (i.e. non FFT) parts of the SETI app. That is completely open at this point.

---

Hows development going?

~BoB

---

Do you Good Search for Seti@Home? http://www.goodsearch.com/?charityid=888957
Or Good Shop? http://www.goodshop.com/?charityid=888957

I now have a new PS3 binary available (same download link as before). This binary was built from the modified 2.3A sources from lunatics.at. There appear to be some modest improvements in performance. Please try it out. Thanks.

---

I have a noob question here, if/when this does succeed, it will be easier to install on the ps3 ..right? Sounds like a nightmare right now, Hell even getting Linux on there sounds like it (..never touched linux.)

But it sounds like you guys are on to something, yall just need to work together and not have just 1 guy doing it o.0

tbh it's going to be a very niche thing without getting help from Sony to put it in the XMB, but I doubt they'll let that happen due to the open source nature of the project..

I doubt you'll get any real performance out of seti@home on PS3 Linux unless you completely recode things from the ground up to take advantage of the SPE's and all the other clever things the Cell can do.

Bare in mind that the chicken sources as good as they are are optimised for current x86 architectures, the Cell is a *completely* different beast..

I've recompiled the chicken sources onto PS3 YDL before, but it was so slow as to be pointless without a ground-up recode..

---

I'll work on it during my holidays in July.
Gaurav, could you please send my sources on which you're working. I sent my e-mail address by private message.

Thanks,
ThierryH

---