Message boards : Graphics cards (GPUs) : Aborted by project... redundant result
| Author | Message |
|---|---|
|
First one of these I have seen- | |
ID: 4953 | Rating: 0 | rate:
 /   Reply
Quote
| |
 Paul D. BuckSend message Joined: 9 Jun 08 Posts: 1050 Credit: 37,321,185 RAC: 0 Level  Scientific publications      | |
First one of these I have seen- I hate that too ... This is another case where I think the credit policy designed way back when does not make as much sense as I would like. I know it it cannot be used for anything ... but, it is the currency by which we gain acknoledgement of our contributions ... to lose a task through no fault of your own stinks. ONe of the reasons I like CPDN, you get paid for trickles as you complete phases of work so that even if it dies ... well, you get paid for the effort to the last "payday" (Trickle). This is one of the reasons I stopped doing work for Sztaki and am not enthused about Orbit even though Orbit, for one, is the type of project I really want to support. Heck, I am only here because I want to use my GPU and to exercise this option and have no enthusiasm for supporting SaH ... for too many reasons to go into here ... | |
| ID: 4955 | Rating: 0 | rate:
/ Reply
Quote
| |
|
The workunits are resend some hours bevore the deadline is over? | |
| ID: 4957 | Rating: 0 | rate:
/ Reply
Quote
| |
|
Paul D. Buck Send message Joined: 9 Jun 08 Posts: 1050 Credit: 37,321,185 RAC: 0 Level Scientific publications | |
The workunits are resend some hours bevore the deadline is over? No ... it shouldn't ... HOWEVER, Let us say that the task does not come in by the deadline ... the work gets re-issued ... THEN, in this case, 7 hours late, the task is returned by the first participant. The second participant gets, ahem, cheated ... I know that the project can fiddle with the settings to minimize this, like changing the speed at which the tasks are re-issued so that there is more of a delay. Again, this is a place where the interests of the participant community are not taken into account well by the BOINC System design. I think most of us are willing to accept a certain "loss rate" of tasks that don't complete for one reason or another ... but ... in this case it was, well, one of those losses that seems a little unfair ... This can be especially annoying in that BOINC is designed to not contact the server that often and so, he started the task and wasted time on a task that was canceled. | |
| ID: 4960 | Rating: 0 | rate:
/ Reply
Quote
| |
|
ExtraTerrestrial Apes Volunteer moderator Volunteer tester  Send message Joined: 17 Aug 08 Posts: 2705 Credit: 1,311,122,549 RAC: 0 Level  Scientific publications  | |
|
As Paul already said, what we're seeing here is the following: someone went past the deadline, the WU got reissued and the first guy returned the WU first and got credit for it. The WU was aborted by the server on JAMCs PC. So far everything is alright - it's better to waste 10h of work than to let JAMC crunch along and let him waste e.g. 16h for the full WU. | |
| ID: 4998 | Rating: 0 | rate:
/ Reply
Quote
| |
Beyond Send message Joined: 23 Nov 08 Posts: 1112 Credit: 6,162,416,256 RAC: 0 Level  Scientific publications | |
|
This is really pretty simple. If you issue a WU to someone and they crunch it in good faith they need to get "paid". You do NOT need to "pay" the user who returned it late as they broke the contract you had with them, that's up to you. Simple contract law, if you want to look at it in legal terms. If you want to look at it in terms of good will, you will destroy your relationship with your users if you fail to "pay" them for work you asked them to do and they worked on in a timely manner. IMO a duplicate WU here and there isn't going to kill you. Don't allow the server to cancel WUs that have already been started, and allow credit for duplicate WUs. | |
| ID: 5002 | Rating: 0 | rate:
/ Reply
Quote
| |
|
ExtraTerrestrial Apes Volunteer moderator Volunteer tester Send message Joined: 17 Aug 08 Posts: 2705 Credit: 1,311,122,549 RAC: 0 Level Scientific publications | |
IMO a duplicate WU here and there isn't going to kill you. Don't allow the server to cancel WUs that have already been started, and allow credit for duplicate WUs. You'd trade off performance for better user relations. It's not a bad solution, but it's not ideal either. I still prefer my solution - if it can be implemented without too much hassle. MrS ____________ Scanning for our furry friends since Jan 2002 | |
| ID: 5009 | Rating: 0 | rate:
/ Reply
Quote
| |
|
Beyond Send message Joined: 23 Nov 08 Posts: 1112 Credit: 6,162,416,256 RAC: 0 Level Scientific publications | |
What is not correct is that JAMC doesn't get credit for the time he already spent. Could help to create a new rule: It sounded difficult to do but if you CAN do it without too much trouble it's a good solution IMO. | |
| ID: 5010 | Rating: 0 | rate:
/ Reply
Quote
| |
|
Paul D. Buck Send message Joined: 9 Jun 08 Posts: 1050 Credit: 37,321,185 RAC: 0 Level Scientific publications | |
As Paul already said, what we're seeing here is the following: someone went past the deadline, the WU got reissued and the first guy returned the WU first and got credit for it. The WU was aborted by the server on JAMCs PC. So far everything is alright - it's better to waste 10h of work than to let JAMC crunch along and let him waste e.g. 16h for the full WU. I think a rule that can be implemented on the server without too much trouble though I am not sure if the canceled result record contains the information, no, it does not appear to have the run time saved ... Anyway, my feeling is that we don't have to get too complex here, a partial payment, even if it is only 25% of the norm for work started but not needed is a token of good faith. The problem is if it can be determined if the task has been started by the participant. What I see in the database does not indicate that the information is preserved. As you say, it should be ... but probably isn't. This is one of those areas where I have been yelling at the ceiling for years ... and writing about. The developers and in many cases the project staffs pay no attention to the interests and desires of the participant community on which BOINC depends. Which, I feel, is one of the reasons that the number of participants has not really been growing that much. When I stopped doing BOINC about three years ago we had a million something participants ... we have 1.5 million roughly at this time. A very anemic growth ... with the number of long term projects rising, well, even with the increases in speed of processing the ability to get the work done is not really improving ... Anyway, I am not saying that this is the case here ... just that the design of BOINC assumes that the participant will take all manner of abuse and that the risk / reward calculus is meaningless and that participants love abuse ... Forgive the rant ... And, again, I am not saying that GPU Grid is in anyway the owner of this issue ... I think it is an inherent flaw in the design decisions made while finalizing BOINC ... note that the expected average runtime for tasks when these decisions were made was 2-6 hours ... with 2 hours being more the norm... I have been noticing that most work now issued is 6-40 hours with at least 3 projects issuing tasks that have one to two hundred hours of run time as the norm and only one of those uses the "trickle" system which was intended for tasks with runtimes in this realm ... | |
| ID: 5017 | Rating: 0 | rate:
/ Reply
Quote
| |
|
I agree with you Paul about the "time served" issue. I've have very few WU's that I've gotten no credit for. but I am totally in concurrence. | |
| ID: 5022 | Rating: 0 | rate:
/ Reply
Quote
| |
Stefan Ledwina Send message Joined: 16 Jul 07 Posts: 464 Credit: 135,911,881 RAC: 101 Level  Scientific publications | |
|
Hmm.. I always thought if the server sends a request to BOINC to delete a redundant result, BOINC would only delete it if it hasn't started crunching. Otherwise it would run it to completition and the user should also get the Credit for it. | |
| ID: 5023 | Rating: 0 | rate:
/ Reply
Quote
| |
|
ExtraTerrestrial Apes Volunteer moderator Volunteer tester Send message Joined: 17 Aug 08 Posts: 2705 Credit: 1,311,122,549 RAC: 0 Level Scientific publications | |
Don't know if it was changed or it is maybe a bug? I could imagine it's a setting in the BOINC software. GPU-Grid is speed / latency limited, so they want results back fast.. that's why I suppose cancelling a WU which was already started is by design and not a bug. MrS ____________ Scanning for our furry friends since Jan 2002 | |
| ID: 5025 | Rating: 0 | rate:
/ Reply
Quote
| |
|
Paul D. Buck Send message Joined: 9 Jun 08 Posts: 1050 Credit: 37,321,185 RAC: 0 Level Scientific publications | |
I agree with you Paul about the "time served" issue. I've have very few WU's that I've gotten no credit for. but I am totally in concurrence. Pat, Yes your basic math is correct. BUT, the models have increased in complexity and run times. My newest rig is i7 based. * Virtual CPUs and each of those (L3 Cache? CLock?) is more capable than my not that old Q9300 ... regardless ... were I able to get the preference page to accept changes I would be able to show my signature showing that I have nearly 40 projects currently going. Stefan just below has 38 inactive and 24 active ... On my tracking chart I have 70+ to which I have contributed, are active, or to which my benchmark participant has contributed to in their history. What I am trying to say is that the demand, I think, is growing faster than the productivity curve of us geeks updating our systems. Though I do grant you that the power of these systems is such that astounding leaps in CS score are possible. I have made over 30K here at GPU Grid and my GPU is nothing special at all as these things go. I was looking at my April printout and since then I have nearly doubled (for example) my Rosetta contribution and I have not had them even as a priority. Still, by basic point remains. BOINC was intended to capture idle CPU of large masses of computers and it has not done that. Heck, nearly half, if memory serves, of the SaH CLassic folks did not make the transition to BOINC ... And given the commentary of many in the SaH forums I am not so sure how common the adoption of GPU computing will be ... will it be only a sub-set of us geeks, a large sub-set, what? No one knows for sure ... I am wild about it for a couple reasons, one is that my wife is paring down my farm from 10 to who knows where she will stop ... (at 5 right now) ... so this gives me a way to increase computing power and tasks in flight over base machines, and the ability to gradually increase processing power over time on an incremental basis without having to scrap whole machines, or the major core of them. | |
| ID: 5026 | Rating: 0 | rate:
/ Reply
Quote
| |
|
Paul D. Buck Send message Joined: 9 Jun 08 Posts: 1050 Credit: 37,321,185 RAC: 0 Level Scientific publications | |
Don't know if it was changed or it is maybe a bug? It may not be a bug to the project, but it may be a bug to the participant ... :) | |
| ID: 5027 | Rating: 0 | rate:
/ Reply
Quote
| |
|
ExtraTerrestrial Apes Volunteer moderator Volunteer tester Send message Joined: 17 Aug 08 Posts: 2705 Credit: 1,311,122,549 RAC: 0 Level Scientific publications | |
It may not be a bug to the project, but it may be a bug to the participant ... :) It's a feature not perfectly implemented yet ;) MrS ____________ Scanning for our furry friends since Jan 2002 | |
| ID: 5033 | Rating: 0 | rate:
/ Reply
Quote
| |
Message boards : Graphics cards (GPUs) : Aborted by project... redundant result
