| Index | Recent Threads | Unanswered Threads | Who's Active | Guidelines | Search |
| World Community Grid Forums
|
| No member browsing this thread |
|
Thread Status: Active Total posts in this thread: 33
|
|
| Author |
|
|
Speedy51
Veteran Cruncher New Zealand Joined: Nov 4, 2005 Post Count: 1326 Status: Offline Project Badges:
|
I agree with you if you are wanting to return results as fast as possible. My thoughts are with the limited amount of work there seems to be at the moment the quicker we return it the quicker we may run out of work It is not a question of returning results as soon as possible, but of trying to run the work units where they are the most efficient. I would like to deploy my electricity where it does the most work per watt. Cheers +1 From my point of view you get more work per watt from Windows for SCC and more work from MCM using Linux because those other platforms that the project take longest to run on. We are both right with what we are saying. You are not going to change my view and I am not going to change your view. :-) Linux is not for everyone. I am guessing no one is running a Ryzen 9 7900 X if you are concerned about watts out of interest what do you do with your watts when you run out of work. I either wait until work starts coming in all go to another project ![]() [Edit 2 times, last edit by Speedy51 at Aug 17, 2023 4:38:25 AM] |
||
|
|
MJH333
Senior Cruncher England Joined: Apr 3, 2021 Post Count: 301 Status: Offline Project Badges:
|
AgrFan
Right now, it's best to run MCM on Windows and SCC on Linux using separate device profiles. SCC uses VINA which runs quicker on Linux than Windows. Thank you for reminding me about this. I’ve recently been running SCC on my Windows 10 Zen 3 machine and decided to try running it on Linux instead. After quite a bit of failing to get it to work (!), I’ve now managed to get SCC to run in Ubuntu using WSL2. The results are astonishing. In Windows an SCC task would generally take 2.5 to 3 hours (with some outliers). Running on the same machine in WSL2, the tasks are generally taking about 45 minutes. Wow!Cheers, Mark |
||
|
|
BobbyB
Veteran Cruncher Canada Joined: Apr 25, 2020 Post Count: 638 Status: Offline Project Badges:
|
Yes a good reminder.
----------------------------------------When WCG started having problems a week or two ago, I removed the limits for MCM on my Ubuntu's and took what there was. Have now reset them to what they were. I will keep my eye on them. 6 hours later: Yeah they are doing it OK. I only have two Win 4core laptops so 1 SCC But tomorrow is the start of the weekend and we know what that means. what do you do with your watts when you run out of work. I either wait until work starts coming in all go to another project Same here and yes am not too concerned with watts. Where I am all watts are 99.9% hydro.[Edit 4 times, last edit by BobbyB at Aug 17, 2023 10:41:51 PM] |
||
|
|
cubes
World Community Grid Tech, Mapping Cancer Markers and Help Conquer Scientist Canada Joined: Mar 3, 2007 Post Count: 58 Status: Offline Project Badges:
|
Interesting finding -- thank you for sharing it.
For all MCM workunits these days, the bulk of the compute time is spent inside calls to the SvmLightLib API, which wraps SVM-Light. Multiple points in the SVM-Light code call the C standard library function clock(), which presumably ends up as times() system calls. SVM training and testing is an iterative process, and different workunits will vary in the number of signatures evaluated, the number of iterations per model, and the time required per iteration, so the number of system calls and relative time spent in them will vary. The numbers are generally higher for the current dataset, but 35% does seem especially high. Note: MCM is a unique use-case of SvmLightLib. Any problems with our application of SVM-Light/SvmLightLib to MCM are our problems, not those of the packages' authors. I recall the WCGrid technical team mentioning paying some attention to those clock calls and trying to reduce their number in the MCM application, back when it was being adapted to WCGrid. I was strictly on the Science side of the wall back then, so couldn't see any details. No longer the case now. I'll open a ticket about the issue and look into it when time allows. |
||
|
|
AgrFan
Senior Cruncher USA Joined: Apr 17, 2008 Post Count: 396 Status: Offline Project Badges:
|
I recall the WCGrid technical team mentioning paying some attention to those clock calls and trying to reduce their number in the MCM application, back when it was being adapted to WCGrid. I was strictly on the Science side of the wall back then, so couldn't see any details. No longer the case now. I'll open a ticket about the issue and look into it when time allows. cubes, Thank you for taking the time to respond. This is great news. I've stopped running MCM on Ubuntu due to the long-elapsed times vs. Windows. You may want to focus on the NFCV units first since they are egregiously slow on Linux. Please keep us posted on any further developments. Thanks again. Agrfan
[Edit 8 times, last edit by AgrFan at Aug 17, 2023 8:24:04 PM] |
||
|
|
thunder7
Senior Cruncher Netherlands Joined: Mar 6, 2013 Post Count: 238 Status: Offline Project Badges:
|
Interesting finding -- thank you for sharing it. ... I recall the WCGrid technical team mentioning paying some attention to those clock calls and trying to reduce their number in the MCM application, back when it was being adapted to WCGrid. I was strictly on the Science side of the wall back then, so couldn't see any details. No longer the case now. I'll open a ticket about the issue and look into it when time allows. For the record, this is what I see on a MCM task on linux (currently @ 26.5% system time, while running a MCM and SCC mix): sudo perf record -p <pid> wait 10 seconds ^C sudo perf report Samples: 67K of event 'cycles', Event count (approx.): 46184408072▒ 0.51% wcgrid_mcm1_map [kernel.kallsyms] [k] update_min_vruntime [Edit 1 times, last edit by thunder7 at Aug 17, 2023 7:32:41 PM] |
||
|
|
Sgt.Joe
Ace Cruncher USA Joined: Jul 4, 2006 Post Count: 7846 Status: Offline Project Badges:
|
out of interest what do you do with your watts when you run out of work When I run out of work and the backup project has no work, I turn off the machines. When they are off they use zero watts. Cheers
Sgt. Joe
*Minnesota Crunchers* |
||
|
|
Speedy51
Veteran Cruncher New Zealand Joined: Nov 4, 2005 Post Count: 1326 Status: Offline Project Badges:
|
out of interest what do you do with your watts when you run out of work When I run out of work and the backup project has no work, I turn off the machines. When they are off they use zero watts. Cheers The only way I am aware of to know they will use 0 watts for sure is if you unplug PC/screen from the wall. ![]() |
||
|
|
Sgt.Joe
Ace Cruncher USA Joined: Jul 4, 2006 Post Count: 7846 Status: Offline Project Badges:
|
out of interest what do you do with your watts when you run out of work When I run out of work and the backup project has no work, I turn off the machines. When they are off they use zero watts. Cheers The only way I am aware of to know they will use 0 watts for sure is if you unplug PC/screen from the wall. If I have them turned off, I unplug them. If they are off, there is no need to have them plugged in. Cheers
Sgt. Joe
*Minnesota Crunchers* |
||
|
|
BobbyB
Veteran Cruncher Canada Joined: Apr 25, 2020 Post Count: 638 Status: Offline Project Badges:
|
Is there really anyone here worried about watts except as discussed above. MCM better on Windows and SCC better on Linux.
I appreciate that co$t is a factor in some places and I bet in these cases they have cut back on time and/or machines. If watts is THE factor then you are probably not doing this anymore and are not reading this post. The fact we are here indicates we made the conscious decision that the project is worth it along with the backup projects also. |
||
|
|
|