Longer MB units

Message boards : Number crunching : Longer MB units
Message board moderation

To post messages, you must log in.

1 · 2 · Next

AuthorMessage
Profile [B^S] madmac
Volunteer tester
Avatar

Send message
Joined: 9 Feb 04
Posts: 1175
Credit: 4,754,897
RAC: 0
United Kingdom
Message 888327 - Posted: 25 Apr 2009, 20:22:26 UTC

Are we getting longer (timewise) MBs or is it because of the different ranges again. I can normal do a MB in 105 minutes however some are taking around 170 minutes. It seems to be one in every four for me.Can someone explain please.
ID: 888327 · Report as offensive
Fred W
Volunteer tester

Send message
Joined: 13 Jun 99
Posts: 2524
Credit: 11,954,210
RAC: 0
United Kingdom
Message 888332 - Posted: 25 Apr 2009, 20:34:29 UTC - in response to Message 888327.  

It all depends on the Angle Range.

F.
ID: 888332 · Report as offensive
Cosmic_Ocean
Avatar

Send message
Joined: 23 Dec 00
Posts: 3027
Credit: 13,516,867
RAC: 13
United States
Message 888334 - Posted: 25 Apr 2009, 20:36:37 UTC

I believe crunch time depends on the angle range, which determines the number of flops, which determines the claimed credit. A 'typical' MB WU has an AR of ~0.445. Some are slightly higher, some are slightly lower, but 0.445 used to be approximately 45 credits, but it appears that has skewed down to around 38 recently.
Linux laptop:
record uptime: 1511d 20h 19m (ended due to the power brick giving-up)
ID: 888334 · Report as offensive
Profile ML1
Volunteer moderator
Volunteer tester

Send message
Joined: 25 Nov 01
Posts: 21135
Credit: 7,508,002
RAC: 20
United Kingdom
Message 888341 - Posted: 25 Apr 2009, 21:27:00 UTC
Last modified: 25 Apr 2009, 21:27:27 UTC

There's a very large batch of VLARs crawling through at the moment.

Especially long/slow/bad for us CUDA crunchers :-(

Happy crunchin',
Martin
See new freedom: Mageia Linux
Take a look for yourself: Linux Format
The Future is what We all make IT (GPLv3)
ID: 888341 · Report as offensive
Profile zoom3+1=4
Volunteer tester
Avatar

Send message
Joined: 30 Nov 03
Posts: 66302
Credit: 55,293,173
RAC: 49
United States
Message 888347 - Posted: 25 Apr 2009, 21:53:57 UTC - in response to Message 888341.  

There's a very large batch of VLARs crawling through at the moment.

Especially long/slow/bad for us CUDA crunchers :-(

Happy crunchin',
Martin

And they keep on coming, Eek!
Savoir-Faire is everywhere!
The T1 Trust, T1 Class 4-4-4-4 #5550, America's First HST

ID: 888347 · 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 888349 - Posted: 25 Apr 2009, 21:57:28 UTC - in response to Message 888327.  

Are we getting longer (timewise) MBs or is it because of the different ranges again. I can normal do a MB in 105 minutes however some are taking around 170 minutes. It seems to be one in every four for me.Can someone explain please.

Here's a graph of execution time vs. angle range for a sample of CPU hosts,



The shape of the red line is the approximation now being used for initial estimates of how long work will take. As you can see, it is only a rough estimate, but usable for work fetch calculations. The image is from the Estimates and Deadlines Revisited thread, which has more discussion.
                                                            Joe

ID: 888349 · Report as offensive
Ianab
Volunteer tester

Send message
Joined: 11 Jun 08
Posts: 732
Credit: 20,635,586
RAC: 5
New Zealand
Message 888384 - Posted: 26 Apr 2009, 1:50:34 UTC

The longer units take more time, but you get more credits for them.

Instead of the normal 38, you might get 50 to 80. So unless they are snarling up your CUDA rig then they aren't a problem.

Ian
ID: 888384 · Report as offensive
Profile zoom3+1=4
Volunteer tester
Avatar

Send message
Joined: 30 Nov 03
Posts: 66302
Credit: 55,293,173
RAC: 49
United States
Message 888387 - Posted: 26 Apr 2009, 1:58:30 UTC - in response to Message 888384.  

The longer units take more time, but you get more credits for them.

Instead of the normal 38, you might get 50 to 80. So unless they are snarling up your CUDA rig then they aren't a problem.

Ian

Yeah, their just unexpected, But not unwanted, At least to Me. My Cache was at 41 days earlier today, Now It's at just over 30 days. Inflation has come to Seti.
Savoir-Faire is everywhere!
The T1 Trust, T1 Class 4-4-4-4 #5550, America's First HST

ID: 888387 · Report as offensive
Terror Australis
Volunteer tester

Send message
Joined: 14 Feb 04
Posts: 1817
Credit: 262,693,308
RAC: 44
Australia
Message 888388 - Posted: 26 Apr 2009, 2:16:36 UTC

Been getting almost nothing but VLAR's for 3 days now, and the ones from the 26Dec08 series are particularly nasty and prone to hanging.

It's all those VLAR units that have been previously bounced by CUDA users being re-sent because they're clogging up the system. It's the law of "Swings and Roundabouts". *Someone* has to eat all the black jelly beans that have been left behind in the jar :-)

Brodo
ID: 888388 · Report as offensive
Profile zoom3+1=4
Volunteer tester
Avatar

Send message
Joined: 30 Nov 03
Posts: 66302
Credit: 55,293,173
RAC: 49
United States
Message 888394 - Posted: 26 Apr 2009, 3:21:39 UTC - in response to Message 888388.  
Last modified: 26 Apr 2009, 3:28:24 UTC

Been getting almost nothing but VLAR's for 3 days now, and the ones from the 26Dec08 series are particularly nasty and prone to hanging.

It's all those VLAR units that have been previously bounced by CUDA users being re-sent because they're clogging up the system. It's the law of "Swings and Roundabouts". *Someone* has to eat all the black jelly beans that have been left behind in the jar :-)

Brodo

I like almost any flavor but Black Jelly Beans and Black Licorice I abhor the most, The rest are fine, My Nephew likes Black Licorice and I like Red Vine Licorice(Which He detests, go figure).

Oh and some of the VLARs do hang and I've been getting a lot that actually work, But wow, cpu or not they take a long, Long time to do.

Oh and the cache is now down to 20 days, This is too weird.
Savoir-Faire is everywhere!
The T1 Trust, T1 Class 4-4-4-4 #5550, America's First HST

ID: 888394 · Report as offensive
Chelski
Avatar

Send message
Joined: 3 Jan 00
Posts: 121
Credit: 8,979,050
RAC: 0
Malaysia
Message 888402 - Posted: 26 Apr 2009, 4:02:21 UTC

I run CUDA with VLARkill like most people here. Of course there are always fear that those VLAR units will go round and round until they get killed for having errored out too many times. Right now, looking at the wingmen my MBs validate against, chances of a CUDA-CUDA pairing seems to be around 10% (anyone got a better statistics?)

If a unit has about 5 resends for validation, chances of being rejected due to CUDA-CUDA pairing for everytime it gets resend is approx 1 in a million (2 initial replications + 5 resends, could have 1 CPU there and if the other 6 are CUDA vlarkills then it will die). (i'm not expert in how it works, this is the impression in my mind, plesae kindly correct if its wrong.. Of course this assumes that all CUDA clients are vkill ones

Hopefully the rebranding script becomes easy enough to run (or integrated into the opt app) before CUDA-CUDA pairings becomes higher
ID: 888402 · Report as offensive
Profile bloodrain
Volunteer tester
Avatar

Send message
Joined: 8 Dec 08
Posts: 231
Credit: 28,112,547
RAC: 1
Antarctica
Message 888407 - Posted: 26 Apr 2009, 5:23:33 UTC - in response to Message 888402.  

yeah i notice mine going up to 31 hours on 2 of them and i am running a extreme oc 9800gtx+

but what really gets me are my is the wingmen's that slow it down.
ID: 888407 · Report as offensive
Speedy
Volunteer tester
Avatar

Send message
Joined: 26 Jun 04
Posts: 1643
Credit: 12,921,799
RAC: 89
New Zealand
Message 888408 - Posted: 26 Apr 2009, 5:34:03 UTC - in response to Message 888394.  
Last modified: 26 Apr 2009, 5:53:49 UTC

Been getting almost nothing but VLAR's for 3 days now, and the ones from the 26Dec08 series are particularly nasty and prone to hanging.

As of 26 Apr 2009 5:20:09 UTC 26Dec08 has done 10 channels so it has 4 channels left to go. I've had nothing but AP units for the last 8 days my pending credit is 19,973.40, i have all aps selected to get work from. all my pending credit is from 16 ap units. My RAC is only 98.29. Should it be higher than that with a pending credit of 19,973.40? Thank's in advance.
ID: 888408 · Report as offensive
Profile zoom3+1=4
Volunteer tester
Avatar

Send message
Joined: 30 Nov 03
Posts: 66302
Credit: 55,293,173
RAC: 49
United States
Message 888410 - Posted: 26 Apr 2009, 6:09:47 UTC

For the moment the cache is down to a little over 12 days and the times are more like normal. Oh well, It could be worse I guess the PC could get the swine flu. ;)
Savoir-Faire is everywhere!
The T1 Trust, T1 Class 4-4-4-4 #5550, America's First HST

ID: 888410 · Report as offensive
Speedy
Volunteer tester
Avatar

Send message
Joined: 26 Jun 04
Posts: 1643
Credit: 12,921,799
RAC: 89
New Zealand
Message 888413 - Posted: 26 Apr 2009, 6:20:57 UTC - in response to Message 888410.  

For the moment the cache is down to a little over 12 days and the times are more like normal. Oh well, It could be worse I guess the PC could get the swine flu. ;)

You are so right regarding Swine flu & thank's for info
ID: 888413 · Report as offensive
Profile zoom3+1=4
Volunteer tester
Avatar

Send message
Joined: 30 Nov 03
Posts: 66302
Credit: 55,293,173
RAC: 49
United States
Message 888415 - Posted: 26 Apr 2009, 6:29:45 UTC - in response to Message 888413.  
Last modified: 26 Apr 2009, 6:30:44 UTC

For the moment the cache is down to a little over 12 days and the times are more like normal. Oh well, It could be worse I guess the PC could get the swine flu. ;)

You are so right regarding Swine flu & thanks for info

No problem, I last heard of the swine flu was back in 1976. Oh and a PC can't get the swine flu as It isn't biological, Humans are, PCs aren't.

As to WU's I'm having a WU(cuda) just sit there twiddling It's proverbial thumbs from time to time, most of the time suspending the WU and unsuspending the WU starts It back up, Although sometimes I have to shut down Boinc 6.6.20 and restart Boinc to unfreeze the WU, weird indeed.
Savoir-Faire is everywhere!
The T1 Trust, T1 Class 4-4-4-4 #5550, America's First HST

ID: 888415 · Report as offensive
Speedy
Volunteer tester
Avatar

Send message
Joined: 26 Jun 04
Posts: 1643
Credit: 12,921,799
RAC: 89
New Zealand
Message 888726 - Posted: 27 Apr 2009, 8:43:14 UTC - in response to Message 888408.  
Last modified: 27 Apr 2009, 8:44:20 UTC

Been getting almost nothing but VLAR's for 3 days now, and the ones from the 26Dec08 series are particularly nasty and prone to hanging.

Tape 26dec08 has finished. I'd be interested to know the next tape that contains VLAR'S work units is. At present I'm getting all ap units. This has shot my pending credit to 25,012.85. I like how my pending credit is growing, I wish my RAC would start to climb it's been at 98.29 for over two days. I'm sure it will start to increase.
ID: 888726 · Report as offensive
Richard Haselgrove Project Donor
Volunteer tester

Send message
Joined: 4 Jul 99
Posts: 14679
Credit: 200,643,578
RAC: 874
United Kingdom
Message 890296 - Posted: 1 May 2009, 21:59:56 UTC

Looking at the Arecibo observing schedule for December, it looks as if all those VLARs came from P2030: An ALFA Pulsar Survey of the Galactic Plane. Surveys usually give higher ARs, but it looks like they finished the first scan in 2005 and they're now studying individual locations in more detail.

P2030 is managed by Cornell University, and there's a new Cornell publication Arecibo pulsar survey is using 50,000 PCs worldwide to generate supercomputing power linked from the BOINC home page. It's a BOINC project, but under the wing of Einstein@Home, not us.

So we have an interesting BOINC dilemma. Einstein's new ABP1 application relies on precisely the same data that drags SETI/CUDA to its knees. What's a cruncher to do?

Starting today, P2030 is getting five more observing slots on consecutive days (01 May - 05 May). I hope Eric can come up with some way of alleviating the CUDA VLAR problem before those tapes hit the splitters.
ID: 890296 · Report as offensive
Profile Fred J. Verster
Volunteer tester
Avatar

Send message
Joined: 21 Apr 04
Posts: 3252
Credit: 31,903,643
RAC: 0
Netherlands
Message 890304 - Posted: 1 May 2009, 22:25:29 UTC - in response to Message 890296.  
Last modified: 1 May 2009, 22:40:21 UTC

Hi, last time I mainly crunch , AP units, on the 3 QUADs that is.
On my 'old'P4 Dual (2.8GHz), I get MB's. I've taken my LT, out.
.I let it crunch for more then a year, with optimized MB SSE3 applications, once recued by my insurance {:)}.

I do constantly need it for other, quite demanding programs and a LINUX setup.

I get more AP then MB WU's now a day, I'm pleased with them.
2 a day, per core on the fastest QUAD and 3 per 2 day's on the Q6600's.
And, they do are part of the sky survey just like the Multi Beam (Narrow Band;1.42x.xxx.xxx Hz.)!

That means more AR (0.2-0.5) or VLAR?
Still aproblem for CUDA?

Keep On Crunching
ID: 890304 · Report as offensive
Speedy
Volunteer tester
Avatar

Send message
Joined: 26 Jun 04
Posts: 1643
Credit: 12,921,799
RAC: 89
New Zealand
Message 890338 - Posted: 1 May 2009, 23:40:22 UTC - in response to Message 890296.  


Starting today, P2030 is getting five more observing slots on consecutive days (01 May - 05 May). I hope Eric can come up with some way of alleviating the CUDA VLAR problem before those tapes hit the splitters.

Fred J, out of interest where did you see that they are allowing five more observing slots for P2030 on consecutive days (01 May - 05 May)? Thanks in advance. If the tapes hit the splitters before the 10th Anniversary I doubt that it will get looked at because all focus is been into nitpicker
ID: 890338 · Report as offensive
1 · 2 · Next

Message boards : Number crunching : Longer MB units


 
©2024 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.