Message boards : Server programs : Server-side DCF / Duration Correction Factor
Message board moderation
Author | Message |
---|---|
Send message Joined: 27 Jun 06 Posts: 305 |
The rsc_fpops_est values of some projects are way off. It would sure be a good option for the project admins, if there was a feedback like this : effective_rsc_fpops_est = SELECT workunit.rsc_fpops_est * AVG (host.duration_correction_factor) WHERE host.expavg_credit > 100 well, something like this, where effective_rsc_fpops_est would be what the scheduler actually uses for result delivery. The AVG expression needs to be cached somewhere (maybe on a weekly base) of course instead of computing it for each result. |
Send message Joined: 19 Jan 07 Posts: 1179 |
There is already a flops estimation page on the admin interface. What needs to be solved is the general cluelessness of project admins, who may not even know the estimate is supposed to be accurate and supposed to need adjusting over time... [20:02] <Nicolas> I think projects should be given more control over that [20:03] <Nicolas> they *can* know whether they will have work within the next 24 hours [20:03] <Nicolas> so they could send a 24-hour backoff if they don't, lowering server load and worldwide network congestion :) [20:03] <Nicolas> when close to getting work, give backoffs in a random time range [20:13] <zombie67> that requires admins that understand BOINC better. Will never happen [...] [00:39] <Nicolas> projects with very little work should be using max_wus_per_host anyway [00:40] <zombie67> again, your solution relies on admins knowing what they are doing...or caring [00:40] <zombie67> I want my client to be smarter than they are [00:40] <Nicolas> no, I think it can be solved project-side [00:40] <Nicolas> two things: [00:41] <Nicolas> let admins know that they *have* to care [00:41] <zombie67> can't make them [00:41] <Nicolas> and give them better server-side tools! [00:41] <zombie67> won't take the time to learn them [00:41] <zombie67> or even upgrade their server to use them [00:41] <Nicolas> if an admin knows there will be work in 24 hours, we can't expect him to manually lower the backoff by an hour every hour [00:42] <Nicolas> it's like why some projects rarely post front page news: it's a pain to do so [00:42] <zombie67> your premise assumes an admin that knows BOINC, and cares about how it treats the volunteers [00:43] <Nicolas> if an admin doesn't care about treating its volunteers well, why would I volunteer? :) [00:43] <zombie67> it's not the well-run projects that are the problem [00:44] <zombie67> how many projects have problems getting email notifications to work? All of them? [00:44] <zombie67> how many projects have to be told about credit issues by the volunteers? All of them? [00:45] <zombie67> how many projects have to be told to turn on rac decay? all of them? [00:45] <zombie67> the list goes on and on [00:45] <zombie67> opt apps? [00:45] <zombie67> HR [00:45] <zombie67> OSX [00:45] <zombie67> 64 bit [00:45] <zombie67> ... |
Send message Joined: 27 Jun 06 Posts: 305 |
Thanks for the info, I didn't know that. |
Copyright © 2024 University of California.
Permission is granted to copy, distribute and/or modify this document
under the terms of the GNU Free Documentation License,
Version 1.2 or any later version published by the Free Software Foundation.