Message boards : Server and website : GPUGrid sends TOO MANY WUs at a time
Author | Message |
---|---|
Sys Admin: I don't like having WUs queued up on my computers when someone else could be running them. I keep shortening my queue preferences and I'm down to 0.2/0.2 and still you send me too many WUs at a time. | |
ID: 51779 | Rating: 0 | rate: / Reply Quote | |
Projects don't send work (it wouldn't get through your firewall): your client requests work. | |
ID: 51782 | Rating: 0 | rate: / Reply Quote | |
The client will request work to fill the queue no matter the length of the task. Tasks will be downloaded until its full and that could be queue time + 1 min or queue time + 10 days. Its all the same. | |
ID: 51856 | Rating: 0 | rate: / Reply Quote | |
Yes I've tried queues ranging from 0.04/0.04 to 0.1/0.1 to 0.25/0.25. The queue I set is irrelevant unless it's large. | |
ID: 51873 | Rating: 0 | rate: / Reply Quote | |
The client will request work to fill the queue no matter the length of the task. Tasks will be downloaded until its full and that could be queue time + 1 min or queue time + 10 days. Its all the same.These first two sentences are not the same. I define a small queue and the server overfills my queue with too many WUs. What the server thinks is full is far more than I need. "...queue time + 1 min or queue time + 10 days." Where does the + 1 min and + 10 days come from??? The queue time comes from my Computing Preferences "Store at least 0.2 days of work" and "Store up to an additional 0.2 days of work." I ask for less than half a day of work and if I have 2 GPUs it sends me 3 to 4 days of work, way too much. It's taking that "at least" phrase way too far. I suspect this is a BOINC design flaw and GPUGrid may not be able to do anything about it. BOINC is predicated on the lowest denominator and does nothing to address those that do the lion's share of the work. ____________ | |
ID: 51874 | Rating: 0 | rate: / Reply Quote | |
It's not just a matter of getting the QRB for a WU. If I have too many of anything it changes BOINC priorities in ways I cannot control. If a bunch of lengthy WUs pile up like Bald Eagles under a windmill (tRUMP lie #11,731) then other shorter WUs may not even be sent. The shorter WUs may have a deadline at a later date. The only tool I have available to me is abortion. Abortions cost me nothing but may drive up server costs due to extra data traffic. | |
ID: 51877 | Rating: 0 | rate: / Reply Quote | |
The client will request work to fill the queue no matter the length of the task. Tasks will be downloaded until its full and that could be queue time + 1 min or queue time + 10 days. Its all the same.These first two sentences are not the same. I define a small queue and the server overfills my queue with too many WUs. It's an example that no matter what the task length is the client on your own PC will fill it with whatever size task is gets. Then it will stop requesting more work. The extra time of that last task could be 1 min over your buffer or 10 days. An example. Setup multiple clients if you want to run one project on one card and something else on a 2nd card. Then you can set 0.1 + 0 days for a card and you'll never get extra work until your current task is nearly complete. | |
ID: 51878 | Rating: 0 | rate: / Reply Quote | |
Using multiple clients to compensate for this deficiency is unacceptable. | |
ID: 51884 | Rating: 0 | rate: / Reply Quote | |
That's your opinion but you've spent more time typing out your complaints about queues you don't understand than the time it would take to setup another client. | |
ID: 51895 | Rating: 0 | rate: / Reply Quote | |
We need a block user feature for people that make ad hominem attacks. | |
ID: 51909 | Rating: 0 | rate: / Reply Quote | |
Message boards : Server and website : GPUGrid sends TOO MANY WUs at a time