 |
|
Questions and Answers :
Windows :
Have I mixed up Tasks / Work units
Message board moderation
| Author | Message |
|---|---|
|
Send message Joined: 28 May 08 Posts: 16 Credit: 32,985 RAC: 0 |
Hello I was happily computing until a few days ago. I forgot that BOINC was running and I shut my computer down. Ouch. I had reached 4% of my task :-(. Next morning I fired up my computer, restarted BOINC, saw him restart computation and forgot about it. But when I checked again a few hours later, I discovered BOINC had decided that my previous task was invalid and had loaded a new one. I felt that losing days of computation was unbearable, so I stopped BOINC, reloaded the last backup and restarted it. What I hadn\'t seen, though, is that the old task/work unit was flagged \"over\"/\"client error\"/\"compute error\" on the server too. Now I have this strange situation: BOINC is trickling but in an invalid task/work unit. Is this any use? Frederic |
 Thyme LawnSend message Joined: 5 Aug 04 Posts: 1283 Credit: 15,824,334 RAC: 0 |
Is this any use? Absolutely. As long as your task continues to return trickles and upload result files the CPDN server will accept them and grant you credits (the page for your task shows a trickle was received last night). "The ultimate test of a moral society is the kind of world that it leaves to its children." - Dietrich Bonhoeffer |
|
Send message Joined: 28 May 08 Posts: 16 Credit: 32,985 RAC: 0 |
Absolutely. As long as your task continues to return trickles and upload result files the CPDN server will accept them and grant you credits (the page for your task shows a trickle was received last night). Thanks. I don\'t care about the credits, I am doing it for the results (as we say in French, we are all in the same boat here). I had seen that the server had received the trickle but I was worried the server might not use my results, which would mean I was burning watts for nothing and wasting calculation time. Thanks for your answer. Frederic |
|
astroWX Send message Joined: 5 Aug 04 Posts: 1496 Credit: 95,522,203 RAC: 0 |
The server will forever keep the Model flagged as \'errored\'. There will be some ugly messages when it finishes, things like the result not being accepted. Ignore them. They\'re boinc messages and exist for other projects; they are meaningless at CPDN. This Project only cares about receiving your work, boinc error messages notwithstanding. Best of luck with it. I don\'t care about the credits A kindred spirit! "We have met the enemy and he is us." -- Pogo Greetings from coastal Washington state, the scenic US Pacific Northwest. |
|
Send message Joined: 28 May 08 Posts: 16 Credit: 32,985 RAC: 0 |
Sorry, but I must say you are probably both wrong. Here is the snipped sequence of messages I got one week ago: 26/07/2008 14:45:20|climateprediction.net|Sending scheduler request: To send trickle-up message. Requesting 0 seconds of work, reporting 0 completed tasks I don\'t know if the way the server handles results from \"Compute error\" tasks changed since both of you checked this or if I had another problem, but I know one thing: if ever a task crashes and BOINC loads a new task, I won\'t try to restore, I\'ll let BOINC work on his new one! One question: is the partially accomplished work of any use to CPDN? Frederic |
|
Send message Joined: 5 Sep 04 Posts: 7629 Credit: 24,240,330 RAC: 0 |
Message from server: Completed result hadcm3istd_0qlf_1920_160_15989616_4 refused: result already reported as error This has been the normal response to the completion of a Coupled Ocean model (also known as a TCM: Transient Coupled Model), after a failure/restore ever since they were released in 2006. It\'s a \"BOINC thing\", and, while it may seem alarming to crunchers, it only really applies to other projects with quorums, and is meaningless on this project, because the project people don\'t rely on BOINC messages to decide which models are OK, and which aren\'t. As has already been said, the data from partially completed models IS of use. It\'s been posted in lots of places many times, as well as in the READMEs, but once again: Data from TCMs is returned once per model year for that year, (in the trickle files), with larger amounts returned every 10 model years by means of a zip file. All of which can tell the project people about the model. But the further a model can run, the more useful the data. Backups: Here |
|
Send message Joined: 28 May 08 Posts: 16 Credit: 32,985 RAC: 0 |
Thanks for the answer. I\'m a bit unsure of what you meant by \"completion\" since Boinc was telling me before switching to the new task that the old one was going to need a few years to complete. But I am happy because - you confirmed that what had been accomplished could of use - the new task has already reached 30% in one week, which means I am pretty sure I will complete it without any problem :-) Frederic |
|
Send message Joined: 5 Sep 04 Posts: 7629 Credit: 24,240,330 RAC: 0 |
On fast desktops, running 24/7, with no other projects, and little other work, the long TCMs can be completed in about 3 months. And there\'s a lot of us doing that. :) |
|
Send message Joined: 28 May 08 Posts: 16 Credit: 32,985 RAC: 0 |
... \"on fast desktops\"! mine is a 2 years old Pentium M. It is my best computer. At that time, dual cores were very new and very hot - physically. The laptop is working 24/7 and it is doing absolutely nothing else, so I don\'t think I can do much better now. The new task is running very fast, I guess it is a very small one, but the previous task was so long I wasn\'t even sure it would be finished by the limit (around 10 years). I believe the limit does not really exist, so I was not worried about that, but frankly, who can tell if cpdn will still be doing something in 10 years? I wonder if the announced estimated duration was not wrong: I started the task on an even older (5 years) and slower laptop. Although the trickles were more frequent after I switched, the estimated duration stayed about the same. Frederic |
mo.v Send message Joined: 29 Sep 04 Posts: 2363 Credit: 14,611,758 RAC: 0 |
Hi Frédéric BOINC\'s estimate of how long a task will take on a particular computer isn\'t always very good, but BOINC keeps revising and correcting its estimate. Your very long 160-year model (the one Thyme Lawn linked to in his above post) certainly wouldn\'t have taken 10 years to complete. Crunching at about 4.5sec/ts a lot of the time you would have completed it before the end of 2009. It\'s definitely possible, and a good idea, to restore crashed models from backup and continue them. But I clicked on + and looked at the stderrout messages on the page for your long model. They include the line Model crashed: umshell1.f: ATM_DYN : NEGATIVE THETA DETECTED. This means your model developed an atmospheric value that\'s impossible in the real world. If this occurs, the models are designed to stop and crash. If you had restored it again, the same problem would most probably have occurred again on the same model date. A few models develop negative pressure and also crash. The cause may be the particular combination of initial parameter values for this model. But the results of the processed years will still be added to the data set for the researchers. If you always want to crunch shorter models you should deselect HADCM in the CPDN preferences section of your account. HADAMs and HADSMs are both relatively short compared with HADCMs. If you disable the screensaver, the models process faster. As you\'re running models on a laptop, it would be a good idea to raise the complete laptop slightly above the table surface - not just the little feet at the back. This allows more air underneath and helps to keep the laptop cool. If members are not sure whether it\'s a good idea to restore a particular model from a backup and continue it, you can ask on the forum. We can look at the reason why it crashed, which usually gives a good indication of whether it can be successfully restored and completed. Cpdn news |
|
Send message Joined: 28 May 08 Posts: 16 Credit: 32,985 RAC: 0 |
Thanks for all these informations! Frederic |
©2024 climateprediction.net
