Welcome to MilkyWay@home

MilkyWay takes a backseat to Einstein ???

Message boards : Number crunching : MilkyWay takes a backseat to Einstein ???
Message board moderation

To post messages, you must log in.

AuthorMessage
Aurum
Avatar

Send message
Joined: 11 Jul 17
Posts: 20
Credit: 1,425,292,869
RAC: 0
Message 68963 - Posted: 11 Aug 2019, 2:31:07 UTC

I like to run MilkyWay with Einstein WUs on the same GPU, a perfect pairing. However, Einstein will continue to request work but MilkyWay stops requesting work and runs out. To get more MilkyWay WUs I have to Suspend Einstein to stock up on MilkyWay.
Is it possible to make MilkyWay as aggressive as Einstein???
ID: 68963 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
swiftmallard
Avatar

Send message
Joined: 18 Jul 09
Posts: 300
Credit: 303,562,776
RAC: 0
Message 68964 - Posted: 11 Aug 2019, 3:03:03 UTC - in response to Message 68963.  

I like to run MilkyWay with Einstein WUs on the same GPU, a perfect pairing. However, Einstein will continue to request work but MilkyWay stops requesting work and runs out. To get more MilkyWay WUs I have to Suspend Einstein to stock up on MilkyWay.
Is it possible to make MilkyWay as aggressive as Einstein???

Are both projects set to the same Resource Share?
ID: 68964 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profile Keith Myers
Avatar

Send message
Joined: 24 Jan 11
Posts: 696
Credit: 540,034,563
RAC: 86,710
Message 68965 - Posted: 11 Aug 2019, 3:40:56 UTC - in response to Message 68963.  

You might try setting a 1million to 1 resource share for MilkyWay:Einstein. Still probably won't work. Einstein will still overload the computer with more work and swamp any request for work from MilkyWay.

The only way I can control Einstein is to get 120 tasks or such and then set NNT. I let MilkyWay grab work continuously with no issues. When Einstein is down under 10 tasks left or so, unset NNT and get another slug of work. Set NNT again.
ID: 68965 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profile mikey
Avatar

Send message
Joined: 8 May 09
Posts: 3315
Credit: 519,946,492
RAC: 22,331
Message 68966 - Posted: 11 Aug 2019, 11:20:32 UTC - in response to Message 68963.  

I like to run MilkyWay with Einstein WUs on the same GPU, a perfect pairing. However, Einstein will continue to request work but MilkyWay stops requesting work and runs out. To get more MilkyWay WUs I have to Suspend Einstein to stock up on MilkyWay.
Is it possible to make MilkyWay as aggressive as Einstein???


I do it by running Einstein on some pc's and MilkyWay on others, no problems.
You will find running the same project on your cpu and gpu will cause problems too as they do not have separate caches.
To fix that I run one project on the cpu and a different project on the gpu, then swap them on another pc.
ID: 68966 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Aurum
Avatar

Send message
Joined: 11 Jul 17
Posts: 20
Credit: 1,425,292,869
RAC: 0
Message 68968 - Posted: 11 Aug 2019, 15:08:20 UTC - in response to Message 68966.  

You will find running the same project on your cpu and gpu will cause problems too as they do not have separate caches.
I did not know that. What cache is that?

I was hoping Dr Jake would just adjust a server setting and put them on a level playing field. But as soon as I posted I realized I'd be told to adjust Resource Share. I never use it since it can have undesired results and besides with multiple computers I'd lose track of which was set to what.

If you watch GPU Load using GPU-z you see that MilkyWay is a short duty cycle while Einstein is medium and GPUGrid is long. By pairing longer projects with MilkyWay it gets a more efficient and complete use of the GPU.
I recently saw a post where someone was using time stamps in the logs to measure the time over multiple WUs for Linux computers to test if pairings produced more points per hour.
ID: 68968 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
wb8ili

Send message
Joined: 18 Jul 10
Posts: 76
Credit: 635,998,708
RAC: 0
Message 68974 - Posted: 14 Aug 2019, 0:37:06 UTC

Aurum -

This problem started near the end of March 2019 after some kind of meltdown of the Milkyway system. In the "News" section look for the topic "30 Workunit Limit Per Request - Fix Implemented" for a discussion.

The issue has never been fixed.

As far as I can tell, it impossible to run both Einstein and Milkyway projects using the same resources (CPU or GPU or both) on the same computer because of the issue you identified.

On the systems where I run Milkyway only, I just let it run out of tasks because Milkyway doesn't fill the cache until it is empty for a couple of minutes. Then it will send me 300 tasks.
ID: 68974 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profile mikey
Avatar

Send message
Joined: 8 May 09
Posts: 3315
Credit: 519,946,492
RAC: 22,331
Message 68975 - Posted: 14 Aug 2019, 10:56:14 UTC - in response to Message 68968.  

You will find running the same project on your cpu and gpu will cause problems too as they do not have separate caches.
I did not know that. What cache is that?

I was hoping Dr Jake would just adjust a server setting and put them on a level playing field. But as soon as I posted I realized I'd be told to adjust Resource Share. I never use it since it can have undesired results and besides with multiple computers I'd lose track of which was set to what.

If you watch GPU Load using GPU-z you see that MilkyWay is a short duty cycle while Einstein is medium and GPUGrid is long. By pairing longer projects with MilkyWay it gets a more efficient and complete use of the GPU.
I recently saw a post where someone was using time stamps in the logs to measure the time over multiple WUs for Linux computers to test if pairings produced more points per hour.


The on in the Boinc Manager that the Boinc Server side software tries to follow for each of us, the one that says Usage Limits under Options, computing preferences, computing. If you have fast gpu workunits and much longer cpu workunits the pc will have troubles keeping both caches filled when you run both kinds of workunits on the same pc.
ID: 68975 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
mmonnin

Send message
Joined: 2 Oct 16
Posts: 162
Credit: 1,004,381,416
RAC: 17,448
Message 68992 - Posted: 28 Aug 2019, 21:19:38 UTC

I always run 1 client for CPUs and a separate client for GPU work on a single PC. That'll fix caching issues between CPU and GPU tasks.
ID: 68992 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profile mikey
Avatar

Send message
Joined: 8 May 09
Posts: 3315
Credit: 519,946,492
RAC: 22,331
Message 68993 - Posted: 29 Aug 2019, 10:48:08 UTC - in response to Message 68974.  

Aurum -

On the systems where I run Milkyway only, I just let it run out of tasks because Milkyway doesn't fill the cache until it is empty for a couple of minutes. Then it will send me 300 tasks.


This is an ongoing problem here at MilkyWay and it started after they upgraded to a newer version of Boinc on the Server. It's a setting someplace that needs changing but no one seems to know which one fixes the problem.
ID: 68993 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote

Message boards : Number crunching : MilkyWay takes a backseat to Einstein ???

©2024 Astroinformatics Group