Questions and Answers : Getting started : credit for missed trickles ?
Message board moderation
Author | Message |
---|---|
Send message Joined: 27 Jun 05 Posts: 74 Credit: 199,198 RAC: 0 |
if I leave my computer disconnected from the net for a week, but still crunching, it is going to miss three or four trickles. Will I get backdated credit when I finally do let it get in touch, or do I have to wait thill the end of the wu? And, while asking, is there any other possible problem if my computer does not check in for several weeks at a time? |
Send message Joined: 5 Sep 04 Posts: 7629 Credit: 24,240,330 RAC: 0 |
You'll get the credit. In a rush. :) No worries about not connecting for a few weeks. There was a problem with trying to upload more than about 35-40 trickles in one go. (The number seemed to vary.) I'm not sure if this has been fixed, but I'm going to use the recent server problem to find out. I've got 26 saved up so far. Should there still be a problem, the cure is simple, and can be found <a> here.</a> |
Send message Joined: 27 Jun 05 Posts: 74 Credit: 199,198 RAC: 0 |
Thanks Les > No worries about not connecting for a few weeks. that's good as one of the advantages of this project is thay you can leave it running unattended for ages - and even without network support. All the other projects would not only run out of work but also refuse to accept it back after a long delay. ;-) |
Send message Joined: 31 Oct 04 Posts: 336 Credit: 3,316,482 RAC: 0 |
Is the 40 trickles problem on server side? I always assumed it must be the BOINC client that causes it, in this case it would not be fixed without a BOINC upgrade. |
Send message Joined: 5 Sep 04 Posts: 7629 Credit: 24,240,330 RAC: 0 |
Ananas I'm going on Thyme Lawn's post <a href="http://www.climateprediction.net/board/viewtopic.php?t=2811"> here,</a> which suggests that it's a scheduler problem. I think I read somewhere that 'some' of these problems had been fixed, but I could be wrong, which is why I'm going to test it. |
Send message Joined: 5 Sep 04 Posts: 7629 Credit: 24,240,330 RAC: 0 |
River~~ It is not a good idea to leave a computer COMPLETELY unattended on this project. Check it at least once a week, (if not daily), because people are always reporting problems. And if it IS to run like this, make sure you disable network access, because, if it fails, it may attempt to download a new model. And again, and again. Etc. Some people have dozens of failed models, apparently because of computer problems. |
Send message Joined: 31 Oct 04 Posts: 336 Credit: 3,316,482 RAC: 0 |
I do run several computers unattended as I don't have a monitor for each PC. What I check very often (usually daily) is : - the results page, so I see if a computer runs wild and keeps crashing models - the hosts page, so I see if a host is nonfunctional, i.e. doesn't deliver trickles within the "usual" time span Having a computer without web access for some time never caused problems for me, a week or two shouldn't hurt at all. |
Send message Joined: 27 Jun 05 Posts: 74 Credit: 199,198 RAC: 0 |
> ...if it IS to run [unattended], make sure you disable network access, because, > if it fails, it may attempt to download a new model. And again, and again. Does this apply to a machine that is alreadyt crunching away happily? Surely if it has got to its first few trickles, even if a wu fails it is not going to go into an infinite download loop? Presumably this project has the daily quota set at 1 wu per cpu? If so, that would limit such damage to 1 per day per cpu. > Some people have dozens of failed models, apparently because of computer > problems. Presumably these WU simply get passed on to other machines? The advice to check the machines every day is clearly good advice, when possible that is what I do so anyway as I am naturally curious to see how my wu on various projects are coming along. The "if not, then at least weekly" is not so helpful (in my particular circumstances) as if I can't check the machines I can't check them, end of story. I occasionally go away from home, with the likelihood of being back in a few days, but the possibility that my return may be delayed for a month or so. So the question I should have asked was: is it better for the project for me to (1) turn off the machine when there is a risk that I may not be back for several weeks, (cost = certain loss of crunching time until I get back, benefit = no risk of chaotic downloads in the event of a failure) (2) leave the machine running cpdn throughout my absence, (cost = small risk of chaotic downloads, benefit = uninterrupted crunching time) I gather from your reply that you'd advise (2), together with preventing net access during my absence. River~~ |
Send message Joined: 5 Aug 04 Posts: 907 Credit: 299,864 RAC: 0 |
trickles can be updated at any time (although I think Thyme Lawn discovered a "limit" of 40 trickles, but that may be fixed by now). If the worst happens and the model crashes and tries to download, there is a limit of 4 per day so that's not too bad. |
©2024 cpdn.org