• Some users have recently had their accounts hijacked. It seems that the now defunct EVGA forums might have compromised your password there and seems many are using the same PW here. We would suggest you UPDATE YOUR PASSWORD and TURN ON 2FA for your account here to further secure it. None of the compromised accounts had 2FA turned on.
    Once you have enabled 2FA, your account will be updated soon to show a badge, letting other members know that you use 2FA to protect your account. This should be beneficial for everyone that uses FSFT.

High Priority BOINC bull donkey.

  • Thread starter Deleted member 88227
  • Start date
D

Deleted member 88227

Guest
Okay, I am tired of this.

Everytime BOINC freaks out about whatever it is that it freaks out about it will stop all running projects to run specific WU's because they're "high priority" and this is freaking ridiculous. Some of these tasks have deadlines for the 11th. That's 10 freaking days from now. I keep a cache of ~1 day at a time. What in the heck is it tripping over?

Now, this isn't that big a deal, right? Wrong. Most of my rigs also have GPUs in them. When BOINC trips out it stops GPU WU's and fills all the CPU slots with these dumb tasks. Even if I have the GPU set to only use .5 CPU it still stops the GPU from running.

Earlier I had 20 WEP tasks running High P on a 10c20t Intel rig. This rig has 2x 980 Ti GPUs with CPU set to .5 per GPU task. It paused one of the GPU tasks to run those stupid WEP tasks. I also have a .25c dedication to WUProp.

So, break down;
1 Collatz (1 GPU .5 CPU)
20 WEB (20 CPU)
1 WUPRop (.25 CPU)

Idle:
1 GPU

How in the crap do I prevent this without just shutting down ALL CPU tasks on my GPU rigs?

I want 1 core dedicated to a GPU so my GPUs run 24/7 without interruption.
 
I run WUProp and Grid computing at 0 and they run fine.

But I do have the same issue if I split everything evenly between 2 projects one will slowly override the other. It's a PITA. Instead of each project getting 1/2 the cores it seems like it just wants to run 1 project on all of them.
 
I've been doing some reading on this and it seems to be an issue with how BOINC requests new tasks.

On projects that run on GPUs I have CPU tasks disabled.

What, from what I understand, is happening is this. The GPUs return WU's so quickly that BOINC starts requesting WU's from other projects just as quickly. Then suddenly when the CPU tasks aren't returning as fast as the GPU tasks it freaks out and loads up the CPU tasks to "catch up" on the extra work it's grabbed.

I'm thinking I can set the network additional work to .1 and .2 respectively to keep tasks in my cache at a minimum and hopefully not trigger BOINC into acting like a fool.

I really wish I could split resource share on a per host basis. Because I would just set all CPU only projects to 0 so they'll only run on the unused CPU cores when available. Since I have some hosts that don't have any CPUs and they run CPU only projects they might get left out.
 
Did alittle reading and it is also how it works scheduling priority. Guess it is based off of how fast it can earn credits and the bigger return the higher the priority for that task.
 
I'm thinking about just setting up Virtual Machines with the remaining cores for CPU only projects. So on my 20 thread rig setup a VM with 18 CPUs and run BOINC on it for the CPU only projects. Then run the GPU projects on the regular OS. Lots of setup though.
 
You beat me with the VM setup; that would have been my suggestion. For CPU crunching that's rather simple.

Plus you could keep those VMs based on needs and start the one or other config
 
Yes..if you are having trouble juggling work units, VM's or multi-clients (harder to set up) are the way to simplify but at a small overhead cost. BOINC has troubles with estimating how much time work units need to complete until it has ran the project some for a while. It will make its best guess and then ask for work based on it. After it has "learned" what your system can do, it should even out. The problem is that if a project sends out work units that have varying sizes, it will essentially end up averaging them.

BOINC will not give you an even split based on your project priority at all times like asked for above. If WCG is set to 100 and PrimeGrid is set to 200. It may very well run PrimeGrid 100% of the time for a while. But then it will shift over to WCG at some point. It will give both projects an equal share of resources OVER TIME based on the resource share you designated. The reason for this is because some work units have sporadic run times like mentioned above. So, BOINC tallies a "resource debt" in which it will try to balance out. When a project has no work available, BOINC will keep a tally on how much it is "owed" once it has work again. There is a cut off limit which I do not recall on the time it "banks". But this is why you will see a newly added project begin filling up work shortly after adding it to your list.
 
Gilthanis said:
So, BOINC tallies a "resource debt" in which it will try to balance out. When a project has no work available, BOINC will keep a tally on how much it is "owed" once it has work again. There is a cut off limit which I do not recall on the time it "banks". But this is why you will see a newly added project begin filling up work shortly after adding it to your list.
Click to expand...

Would setting up a profile (eg home profile) for GPU boxes and assigning the CPU project's priority to zero for each of the home profiles see the same behaviour? i.e. would there still be a resource debt at priority 0?
 
phoenicis, there is always a resource debt if more than one project is added to the client. BOINC will keep tally of how much time you have given to the projects and when it is time to pull work...it will try and level the playing field based on that. At least up to whatever the artificial limit is. The difference is that GPU work should still run if you are dedicating a core for it. I don't know what you are seeing specifically but my GPU's always run regardless of the large number of projects my systems are typically attached to.

I usually only let BOINC have 99% of the CPU's when there is one GPU in the box. 2 GPU's...just depends on projects and my mood.
 
You must log in or register to reply here.
Back
Top