| Index | Recent Threads | Unanswered Threads | Who's Active | Guidelines | Search |
 |
World Community Grid Forums
Category: Support
Forum: Suggestions / Feedback
Thread: Weighted calculation |
| No member browsing this thread |
|
Thread Status: Active Total posts in this thread: 6
|
|
| Author |
|
|
Greg L
Advanced Cruncher Joined: Mar 2, 2007 Post Count: 94 Status: Offline Project Badges: 
|
Something that would be nice to have is the ability to set (roughly) the percentage of your computing time for each project.
----------------------------------------For example there are three active projects at the moment, C19, Cancer, & African Rain. If one could request a spread of say 50% of projects to Cancer, 30% to C19, & 20% to AR it would be useful for allocating assets to where the person wanted them to go. (we'll have to assume for this example that AR units are roughly the same size as the others...) It doesn't have to be exact, but takes into account what the user wants to work on. Say your computer phoned home & needed 10 work units. In the above example it would be 5 Cancer, 3 C19, & 2 AR. However for whatever reason, AR only had one unit available at that particular time, all other things being equal, the one missing work unit would be replaced by a Cancer work unit as that's what the user wants to prioritize. A field in the Device Manager would be a good place to have people provide their input (or instead of percentages, a simple 1-3 ranking of order of preference would work too). (random thought while looking at my stats & thinking "A few more of 'x' instead of 'y' would be nice to have at the moment...")  |
||
|
|
Sgt.Joe
Ace Cruncher USA Joined: Jul 4, 2006 Post Count: 7844 Status: Offline Project Badges: 
|
I think you can come pretty close to what you want by adjusting the number of work units for each project in the custom profiles. It may not be exactly a percentage, but by experimenting with the work unit numbers and the cache setting, you might be pleasantly surprised by the results.
----------------------------------------Cheers
Sgt. Joe
*Minnesota Crunchers* |
||
|
|
Bryn Mawr
Senior Cruncher Joined: Dec 26, 2018 Post Count: 384 Status: Offline Project Badges: 
|
I think you can come pretty close to what you want by adjusting the number of work units for each project in the custom profiles. It may not be exactly a percentage, but by experimenting with the work unit numbers and the cache setting, you might be pleasantly surprised by the results. Cheers The problem with that is that it’s a limit. If you have 10 cores and set 5, 3 and 2 it might go for weeks and be good but then MCM has a blip and suddenly your only processing with half your cores. |
||
|
|
Sgt.Joe
Ace Cruncher USA Joined: Jul 4, 2006 Post Count: 7844 Status: Offline Project Badges:
|
I think you can come pretty close to what you want by adjusting the number of work units for each project in the custom profiles. It may not be exactly a percentage, but by experimenting with the work unit numbers and the cache setting, you might be pleasantly surprised by the results. Cheers The problem with that is that it’s a limit. If you have 10 cores and set 5, 3 and 2 it might go for weeks and be good but then MCM has a blip and suddenly your only processing with half your cores. Well, that could certainly be true, but I would think that even with systems which get rarely looked at, this might work. Even if you only looked at the systems once a week or so, it would not be that difficult to tweak the settings. If this doesn't work, it is only a suggestion. You could always try something different. A percentage setting would be nice though. Cheers
Sgt. Joe
*Minnesota Crunchers* |
||
|
|
Greg L
Advanced Cruncher Joined: Mar 2, 2007 Post Count: 94 Status: Offline Project Badges:
|
I think you can come pretty close to what you want by adjusting the number of work units for each project in the custom profiles. It may not be exactly a percentage, but by experimenting with the work unit numbers and the cache setting, you might be pleasantly surprised by the results. Cheers I've done that before when playing around & seeing that 'this' computer works best with 'that' project & 'that' computer works best with 'this' project. And yes, to a point it works well, my point was that it would be nice to have a way to prioritize things without having to do all that. Of course thinking about it a bit more (slow day around here obviously...), the vast majority of crunchers probably couldn't care less just how things are allocated & those that do are persnickety enough to actually do that deep dive into setting specific computers on specific work profiles to adjust things around. So a low priority project at best. Would be nice though. Anyway, thanks for your (& BM's) response |
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
Weighted....
Assumption for computer and desired share on an 8 core. The below lays out how many you have to specify in the device profile that can be send to the device buffer with an observed device project mean run time. ARP 24 hrs - 12.5% = 24 / 24 * 12.5 of 8 = 1 OPN 3 hrs 50% = 24 / 3 * 50% of 8 = 32 MCM 4 hrs 25% = 24 / 4 * 25% of 8 = 12 No fiddling with app_config as to how many are to be allowed to run concurrent for each science. Oversize the buffer in consideration of run time variability, for instance 1.5 days. This ensures the device will always ask for work when a task completes and keeps asking periodically if something is unavailable, and no idle threads. I'd set ARP to at least N +1 as when a task completes it takes until uploaded and report is complete. Also since the APR task will have been sitting in the queue for a day it will start on the first freed up thread or you could reserve N threads just for ARP in app_config just to have that one running always, back to back. In the verbose version. |
||
|
|
|