canonical result

Message boards : Number crunching : canonical result

To post messages, you must log in.

Previous · 1 · 2 · 3 · Next

AuthorMessage
Profile Paul D. Buck

Send message
Joined: 17 Sep 05
Posts: 815
Credit: 1,812,737
RAC: 0
Message 289 - Posted: 21 Sep 2005, 20:18:18 UTC

The problem is that with windows not tracking time well, it can report all kinds of odd errors. Like the early macintosh computers those versions of windows do not do lots of things the way it should be done.
ID: 289 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
RDC

Send message
Joined: 16 Sep 05
Posts: 43
Credit: 101,644
RAC: 0
Message 294 - Posted: 21 Sep 2005, 21:38:48 UTC

I'm having the same issue with Win XP Home SP2. It's not as severe as what occurs when running on Win98 but it does happen. On WinXP, the CPU bounces between the games and BOINC science application so every few seconds the science application will bounce up to 97% or so and drop back to around 5%. It does drag out the CPU time however. I tested it on a SZTAKI WU and the original estimate at 50% complete stated I should have about 1.5 hours left to crunch (and it's usually pretty close at that point to being somewhat accurate) but took an extra 45 minutes after playing a game for two hours. On the Win98 box, it would have added 2 hours to the crunch time from earlier observations.

I agree with Paul Buck on this one regarding it being how Windows is handling the timing. Luckily I don't use the Win98 box for other applications as much so the problem on that PC can be minimized simply for the reason of not using it for other things as often.
ID: 294 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
keputnam

Send message
Joined: 18 Sep 05
Posts: 24
Credit: 2,088,785
RAC: 0
Message 295 - Posted: 21 Sep 2005, 21:43:33 UTC - in response to Message 250.  

this is definately strange, but i think boinc devs have little impact there. seems like windows is messing something up there. will try to recreate the situation here, have master of orion 3 at home. windows should count the time the process actually got, not the time that past by in real world.
what windows version do you see this with?


Win98 SE and WinXP Home Edition SP2. On both systems, BOINC seems to count elapsed time when a high CPU intensive game is run. I've had similar happen with high CPU intensive graphics programs as well but not to the extent that games have.


The CPU Clock continuing to run while the Science App is left in memory is a Known Problem. The only workaround that I know of is to set Boinc to NOT leave the application in memory when swapping apps. You can set up a different venue for your Win9x machine(s) and let the default stay at remove for W2000/XP etc.

ID: 295 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
RDC

Send message
Joined: 16 Sep 05
Posts: 43
Credit: 101,644
RAC: 0
Message 296 - Posted: 21 Sep 2005, 21:59:25 UTC - in response to Message 295.  

The CPU Clock continuing to run while the Science App is left in memory is a Known Problem. The only workaround that I know of is to set Boinc to NOT leave the application in memory when swapping apps. You can set up a different venue for your Win9x machine(s) and let the default stay at remove for W2000/XP etc.


Yes but that isn't the problem I'm running into as BOINC on the Win98 box is set to crunch only one project, usually Predictor but currently on LHC due to Predictor not issuing WU's until next week. Having BOINC set to dump from memory won't help if I forget to exit out of BOINC before doing something else like play a high CPU intensive game or use a high CPU intensive graphics program on the PC. It's a related issue with the CPU clock timing but not the known and reported issue the devs are aware of.
ID: 296 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
Profile Angus

Send message
Joined: 17 Sep 05
Posts: 412
Credit: 321,053
RAC: 0
Message 345 - Posted: 22 Sep 2005, 23:07:15 UTC

One of our team members has found these two hosts, very similar setups, with vastly different claimed and granted credit, probably from the very different benchmark rataings.

https://boinc.bakerlab.org/rosetta/results.php?hostid=925

https://boinc.bakerlab.org/rosetta/results.php?hostid=921

This should be a good example of why the high/low/middle/average quorum and credit granting system is needed.




Proudly Banned from Predictator@Home and now Cosmology@home as well. Added SETI to the list today. Temporary ban only - so need to work harder :)



"You can't fix stupid" (Ron White)
ID: 345 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
Profile Rebirther
Avatar

Send message
Joined: 17 Sep 05
Posts: 116
Credit: 41,315
RAC: 0
Message 356 - Posted: 23 Sep 2005, 10:07:18 UTC

The problem is that an AMD CPU got 3 times more credits than a Intel CPU. Pls change the outcome WUs set to 3.
ID: 356 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
Profile Angus

Send message
Joined: 17 Sep 05
Posts: 412
Credit: 321,053
RAC: 0
Message 365 - Posted: 23 Sep 2005, 14:36:43 UTC - in response to Message 356.  

The problem is that an AMD CPU got 3 times more credits than a Intel CPU. Pls change the outcome WUs set to 3.


No - they are both AMD Athlon 64 X2 processors, running the same OS

Proudly Banned from Predictator@Home and now Cosmology@home as well. Added SETI to the list today. Temporary ban only - so need to work harder :)



"You can't fix stupid" (Ron White)
ID: 365 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
Profile KWSN Sir Clark

Send message
Joined: 18 Sep 05
Posts: 46
Credit: 387,432
RAC: 0
Message 370 - Posted: 23 Sep 2005, 15:01:09 UTC

One's got a 4200+, the other is a 4400+

One's got 1GB memory and the other's got 512MB.

I would expect different results. The setup's are similar

The benchmark for the higher spec is almost three times that of the slower one.

It must be something to do with the benchmark score. Try manually benchmarking both again and check the background processes that are running which could affect the score.
ID: 370 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
Profile Bok

Send message
Joined: 17 Sep 05
Posts: 54
Credit: 3,514,973
RAC: 0
Message 377 - Posted: 23 Sep 2005, 15:45:21 UTC - in response to Message 370.  
Last modified: 23 Sep 2005, 15:46:36 UTC

One's got a 4200+, the other is a 4400+

One's got 1GB memory and the other's got 512MB.

I would expect different results. The setup's are similar

The benchmark for the higher spec is almost three times that of the slower one.

It must be something to do with the benchmark score. Try manually benchmarking both again and check the background processes that are running which could affect the score.


I would expect different results too, but no way should they be that different.

Not likely to benchmark both, one of them is mine and the other is 'anonymous'.. so unless you can find who anonymous is....The benchmark on the anonymous machine is way higher than anything else out there if you look carefully. More likely it's a recompiled boinc client with some 'added' code I think, or perhaps a bug in that particular client..

Mine is running without anything else on it, stripped down bare and clocked at 2.4Ghz (it's a cruncher only).
Free-DC

Stats for all projects

Custom Stats
ID: 377 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
Profile David E K
Volunteer moderator
Project administrator
Project developer
Project scientist

Send message
Joined: 1 Jul 05
Posts: 1018
Credit: 4,334,829
RAC: 0
Message 378 - Posted: 23 Sep 2005, 15:58:36 UTC

Hi All,

I am going to see how it goes when I get back from vacation, and if we need to start redundancy, we will. The output and number of participants have increased dramatically (THANKS EVERYONE!), so we will still be getting plenty of results to work with.
ID: 378 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
Ocean Archer
Avatar

Send message
Joined: 22 Sep 05
Posts: 32
Credit: 49,302
RAC: 0
Message 389 - Posted: 23 Sep 2005, 19:59:30 UTC

Have a great vacation, David ...
ID: 389 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
pe
Avatar

Send message
Joined: 17 Sep 05
Posts: 4
Credit: 318,774
RAC: 0
Message 406 - Posted: 24 Sep 2005, 5:59:56 UTC

david enjoy your vacation!

if you dont need redundant results and can tell they are valid by one, then why not just give a certain amount of credit to all wu? i am not yet sure if they take around the same time, but that could be an alternative.

greetz, pe.

ID: 406 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
[AF>france>pas-de-calais]symaski62

Send message
Joined: 19 Sep 05
Posts: 47
Credit: 33,871
RAC: 0
Message 409 - Posted: 24 Sep 2005, 8:19:17 UTC
Last modified: 24 Sep 2005, 8:22:02 UTC

https://boinc.bakerlab.org/rosetta/results.php?hostid=1004

^^

ID: 409 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
Profile Webmaster Yoda
Avatar

Send message
Joined: 17 Sep 05
Posts: 161
Credit: 162,253
RAC: 0
Message 618 - Posted: 27 Sep 2005, 14:50:12 UTC - in response to Message 409.  
Last modified: 27 Sep 2005, 14:54:05 UTC

https://boinc.bakerlab.org/rosetta/results.php?hostid=1004


Not sure what you're trying to say here, but if you are giving an example of a host that has strange figures, try this one

https://boinc.bakerlab.org/rosetta/show_host_detail.php?hostid=1711

Look at the measured speed - seems impossible when compared with other hosts with (faster) AMD CPUs listed in the top 10.

Or try this one: https://boinc.bakerlab.org/rosetta/show_host_detail.php?hostid=2242

Measured speed is more than twice that of my own, overclocked Athlon XP 3000+
*** Join BOINC@Australia today ***
ID: 618 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
Pconfig

Send message
Joined: 26 Sep 05
Posts: 6
Credit: 56,254
RAC: 0
Message 653 - Posted: 27 Sep 2005, 19:56:48 UTC - in response to Message 618.  

https://boinc.bakerlab.org/rosetta/results.php?hostid=1004


Not sure what you're trying to say here, but if you are giving an example of a host that has strange figures, try this one

https://boinc.bakerlab.org/rosetta/show_host_detail.php?hostid=1711

Look at the measured speed - seems impossible when compared with other hosts with (faster) AMD CPUs listed in the top 10.

Or try this one: https://boinc.bakerlab.org/rosetta/show_host_detail.php?hostid=2242

Measured speed is more than twice that of my own, overclocked Athlon XP 3000+


Hm, it seems like people are cheating on this or boinc is making mistakes cause this realy is impossible!
Proud member of the Dutch Power Cows
ID: 653 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
Profile mitro

Send message
Joined: 18 Sep 05
Posts: 6
Credit: 225,313
RAC: 0
Message 658 - Posted: 27 Sep 2005, 20:43:55 UTC
Last modified: 27 Sep 2005, 20:47:07 UTC

In case anyone wants a reference point, my dual-core 3800 (which is #2) is running at 2.7GHz so thats why I can pull 700+. A single core 3000+ would have to literally be running 5GHz to have a RAC of 670.

SS of my machine:
http://www.mitro.net/X227P95.gif
ID: 658 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
Profile Webmaster Yoda
Avatar

Send message
Joined: 17 Sep 05
Posts: 161
Credit: 162,253
RAC: 0
Message 669 - Posted: 28 Sep 2005, 1:00:59 UTC - in response to Message 658.  

A single core 3000+ would have to literally be running 5GHz to have a RAC of 670.


Another reference point... My Athlon XP 3000+ (host id 1661)is running at 2340MHz and the best it can do is about 300 credits a day (if left running 24/7 doing nothing else).

The problem with host 1711 seems to be inflated integer and floating point measurements. I'm not saying this participant is cheating (I can't prove it), but there's definitely something wrong somewhere.

I guess from the science point of view it doesn't matter - the work units are valid. But still... Many of us do like the competitive aspect.

Rather than having a quorum, perhaps there could be a system of having benchmark figures for each type of CPU used, based on averages for that type of CPU. Credit could then be capped at say average+20% (to account for overclocked CPUs, faster memory, etc). Anyone claiming more than that would be granted the average only.
*** Join BOINC@Australia today ***
ID: 669 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
Profile Rebirther
Avatar

Send message
Joined: 17 Sep 05
Posts: 116
Credit: 41,315
RAC: 0
Message 3798 - Posted: 21 Nov 2005, 10:03:41 UTC

Take a middle value of 3 results would be fine here because too many users are using an optimized client to get 3 times more credits than other. So if its possible like other projects do it fast, be fair!
ID: 3798 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
Profile dgnuff
Avatar

Send message
Joined: 1 Nov 05
Posts: 350
Credit: 24,773,605
RAC: 0
Message 4105 - Posted: 23 Nov 2005, 23:45:25 UTC - in response to Message 277.  


And, yes, the throw out the high low average mechanism is there to help to accommodate this ...


Dammit Paul, I can't find your post right now. I'm talking about the one where you discussed the idea of actually counting flops.

From what I gather, the infrastructure is already in place in the Boinc client, it's just a case of adding the code to the Rosetta executables, which is rather low on the priority totem pole, since there are other more pressing needs to be addressed by the Bakerlab team.

Here's another interesting thought. Rather than have a quorum per work unit, what changes would be necessary to the Boinc Server back end to have a quorum by architecture. Meaning we assume that all 3.0 GHz P4's will bench at about the same speed.

Two things come of this. Firstly it solves (or it should solve) the Linux / Windows benchmark disparity, and if administered by boinc.berkeley.edu it becomes a database that is project agnostic. Meaning all projects share the same database.

Any Thoughts? Comments? Cream Cakes? Bronx Cheers?
ID: 4105 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
Profile Webmaster Yoda
Avatar

Send message
Joined: 17 Sep 05
Posts: 161
Credit: 162,253
RAC: 0
Message 4106 - Posted: 24 Nov 2005, 0:06:04 UTC - in response to Message 4105.  

Rather than have a quorum per work unit, what changes would be necessary to the Boinc Server back end to have a quorum by architecture. Meaning we assume that all 3.0 GHz P4's will bench at about the same speed.


Not all 3GHz P4's run at the same speed. Just like not all Athlon 64 3700+ run at the same speed. Things like overclocking, RAM speed and size, motherboard etc come into it. The same goes for any processor.
*** Join BOINC@Australia today ***
ID: 4106 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
Previous · 1 · 2 · 3 · Next

Message boards : Number crunching : canonical result



©2024 University of Washington
https://www.bakerlab.org