Bottlenecks of SETI computations

Message boards : Number crunching : Bottlenecks of SETI computations
Message board moderation

To post messages, you must log in.

Previous · 1 · 2

AuthorMessage
DJStarfox

Send message
Joined: 23 May 01
Posts: 1058
Credit: 923,105
RAC: 635
United States
Message 917179 - Posted: 12 Jul 2009, 16:47:45 UTC - in response to Message 916916.  

True, but you'd just have to build it to be gamma ray burst and solar flare resistant. :)

What happens if ET aims a gamma ray burst at it - do you get the detection, or harden against it? :-)


You run like hell because they are trying to destroy us with their gamma rays! LOL
ID: 917179 · Report as offensive
W5GA, W5TAT, W8QR, K6XT

Send message
Joined: 25 Sep 99
Posts: 42
Credit: 21,095,688
RAC: 4,049
United States
Message 917222 - Posted: 12 Jul 2009, 19:23:57 UTC

My own little bottleneck. When Astropulse units were widely available my four machines reached RAC 28,000+. Since Astropulse WU became scarce my RAC has receded to about 15,000. The same machines (3xQuad, 1xcore2duo), the same GPUs, the same hours per day of crunching (7/24), the same KWSN apps. Output has stabilized over the past week in that 15K RAC area.

Now I get the occasional Astropulse WU but 99% are SETI enhanced CPU and GPU WU. The loss of Astropulse WU could be coincidental but the timing is suspicious.

Would the experts call this 'normal' or should I be looking at some error?

Thanks, Art
ID: 917222 · Report as offensive
1mp0£173
Volunteer tester

Send message
Joined: 3 Apr 99
Posts: 8423
Credit: 356,897
RAC: 0
United States
Message 917225 - Posted: 12 Jul 2009, 19:46:37 UTC - in response to Message 917222.  


Now I get the occasional Astropulse WU but 99% are SETI enhanced CPU and GPU WU. The loss of Astropulse WU could be coincidental but the timing is suspicious.

Would the experts call this 'normal' or should I be looking at some error?

Thanks, Art

It suggests that your computer is better at Astropulse than multibeam.

It is a well-known fact that different "angle-range" work units are "high paying" and "low paying" -- and RAC varies depending on the current mix.

It could also be due to a backlog of work to be validated (with a matching "bump" when everything gets caught up).
ID: 917225 · Report as offensive
Josef W. Segur
Volunteer developer
Volunteer tester

Send message
Joined: 30 Oct 99
Posts: 4504
Credit: 1,414,761
RAC: 0
United States
Message 917251 - Posted: 12 Jul 2009, 22:28:16 UTC - in response to Message 917222.  

My own little bottleneck. When Astropulse units were widely available my four machines reached RAC 28,000+. Since Astropulse WU became scarce my RAC has receded to about 15,000. The same machines (3xQuad, 1xcore2duo), the same GPUs, the same hours per day of crunching (7/24), the same KWSN apps. Output has stabilized over the past week in that 15K RAC area.

Now I get the occasional Astropulse WU but 99% are SETI enhanced CPU and GPU WU. The loss of Astropulse WU could be coincidental but the timing is suspicious.

Would the experts call this 'normal' or should I be looking at some error?

Thanks, Art

For Astropulse and Astropulse v5, the optimized apps were maybe 4 times as fast as the stock apps. Since the credit rates adjust based on stock apps, there was a huge credit rate bonus for running optimized AP. The stock Astropulse v505 has some optimizations, so the bonus will be less in future. The server side credit adjustment has reduced the credits for a completed WU to 869.19 for AP_v505 tasks sent July 10, as compared to around 1225 for AP_v5 shortly before the AP_v505 release.

For much of the time before the AP_v505 release the project had an equal number of slots in the Feeder to Scheduler interface for MB and AP work, that made it quite easy to get AP work if you wanted it. But the ap_splitter processes got seriously ahead of the mb_splitter processes, so they switched to a 97:3 balance which approximates the 15872:400 ratio of the number of MB:AP WUs produced by one channel of a 50.2 GB 'tape' file. Until they update the project preferences so Astropulse v505 can be specifically selected, it won't be possible to get many AP WUs.

In short, I call what you're seeing 'normal'.
                                                              Joe
ID: 917251 · Report as offensive
Dan W

Send message
Joined: 25 Jun 09
Posts: 7
Credit: 73
RAC: 0
United Kingdom
Message 917253 - Posted: 12 Jul 2009, 22:37:10 UTC

What might help reduce it somewhat further is if we could get telescopes on the far side of the moon. I'm guessing that the lack of atmosphere there may ease the computational algorithms...(?)

Wouldn't help the computational algorithms, but it would sure help to block the interference from all those cellphones and "I love Lucy".

Which would in turn help ease the work that the algorithm has to do as a guess? Less noise = allowing coarser analysis of the signal for a given quality etc....
ID: 917253 · Report as offensive
W5GA, W5TAT, W8QR, K6XT

Send message
Joined: 25 Sep 99
Posts: 42
Credit: 21,095,688
RAC: 4,049
United States
Message 917294 - Posted: 13 Jul 2009, 4:48:53 UTC - in response to Message 917251.  

Until they update the project preferences so Astropulse v505 can be specifically selected, it won't be possible to get many AP WUs.

In short, I call what you're seeing 'normal'.
                                                              Joe

Josef thank you for the current WU distribution details. Very true about AP WU, I am getting maybe one or two per PC per week at best. Even with credit per AP WU cut in half I still should be running a bit over 20K RAC. But of course that assumes I get some AP WU -- which I ain't!

I saw a thread comment about a script that would somehow get more AP WU through essentially begging (repeated requests) but didn't pursue it, don't know if it works.

Finally, food for (my) thought. I note my group brethren are humming along at about the same RAC they were running a few months ago. Meaning my 50% reduction is unique in the group. It is unknown but statistically improbable that they all decided suddenly to compensate with an upgrade. Sadly I don't know where to go looking. In the spirit of placing telescopes on the far side of the moon, maybe its just my radio station cosmically interfering at HF with the processors!
Regards
Art
ID: 917294 · Report as offensive
Profile Bill Walker
Avatar

Send message
Joined: 4 Sep 99
Posts: 3868
Credit: 2,697,267
RAC: 0
Canada
Message 917365 - Posted: 13 Jul 2009, 15:18:14 UTC - in response to Message 917294.  

I note my group brethren are humming along at about the same RAC they were running a few months ago. Meaning my 50% reduction is unique in the group.

Art


Do you know how many of the others in your group were running optimized aps? My understanding is that stock AP and stock MB give roughly the same credits per hour.

In my case, running optimized aps for AP and MB, I've gone from RAC of 1100 to 1200 a few months ago, when I got almost all AP, down to about 700 with all MB today.

ID: 917365 · Report as offensive
Profile Reuben Gathright
Avatar

Send message
Joined: 8 Mar 01
Posts: 213
Credit: 14,594,579
RAC: 0
United States
Message 917392 - Posted: 13 Jul 2009, 17:50:31 UTC

I wrote a series of reviews on the Intel Socket 775 processors recently. I was amazed to find how much faster the Intel Core 2 Duo's performed in my benchmark testing. Here are some links that may help you with your questions:
Intel Socket 775 Processor Studies, benchmarks and complete analysis from the perspective of a distributed computing participant:
Intel Core 2 Duo E8600 1333Mhz FSB processor with 6mb of L2 Cache
Intel Celeron D 356 3333 Mhz processor with 512Kb of L2 Cache
Intel Pentium 4 630, 3000 Mhz processor with 2Mb of L2 Cache
Intel Pentium D 915, 2800 Mhz processor with 4Mb of L2 Cache
Intel Celeron 352, 3200 Mhz processor with 512Kb of L2 Cache

Overclock with the MSI G31M3-L and Intel E8600 3.33Ghz
Intel D865GLC Socket 478 Motherboard
~How To Overclock The Eee ASUS 1005HA Netbook To 1.9Ghz~
ID: 917392 · Report as offensive
Previous · 1 · 2

Message boards : Number crunching : Bottlenecks of SETI computations


 
©2017 University of California
 
SETI@home and Astropulse are funded by grants from the National Science Foundation, NASA, and donations from SETI@home volunteers. AstroPulse is funded in part by the NSF through grant AST-0307956.