|
1)
Message boards :
Number crunching :
Changes in task names
(Message 751)
Posted 15 Jan 2015 by rhb Post: I noticed the names of tasks have changed recently, while the application is the same. For example, yafu_ali_148122_L121_C118_1421055004_7_0 http://yafu.myfirewall.org/yafu/result.php?resultid=267945 It appears that all the tasks have the _ali_ now. Some of them have just a Cn value, where "n" is presumably the number of digits. Others have an Lm number. I suspected you might have been experimenting with having some of these run to a certain point, then returning the intermediate result to be sent out again as a new WU. It would be interesting to know what changes have been made and why. |
|
2)
Message boards :
Number crunching :
Future status of project
(Message 517)
Posted 15 Dec 2012 by rhb Post: I see the project is temporarily out of work again, and nearly all primes of 112 digits and below in the list have been factored. I wondered if you are considering continuing, perhaps to 120 digits, or if you will just do occasional cleanup runs from now on. I am reluctant to tie up all of my cores at once to yafu for very long runs, but if I could easily select the number of cores (probably two or three of four in my case) I would be happy to crunch regularly up to 120. I have seen runtimes so far up to 6 hours, or 24 cpu hours. If it took 40 cpu hours to factor a very large number, I could choose to factor it in 13 or 20 wall-clock hours. I would be happy to do one every one or two days provided I could run other work at the same time. I wondered if you could informally reserve certain numbers so no one else would be likely to try factoring the ones in your queue. I have learned some interesting things about factoring due to this project. I think you might have more interest if you made it easier to learn about factoring. Specific information such as how many "already factored" issues you have would be beneficial. As far as the deadline scheduling is concerned, I would be happy if I had more control over boinc. I would like to get one yafu task at a time, run it immediately (but swap it out about half the time if it uses all cores), and replace it with another task immediately if it finished quickly. Perhaps the boinc team could be convinced that more flexibility would be useful for other projects also. Thank you for spending time developing this interesting project. |
|
3)
Message boards :
Number crunching :
How much crunching left?
(Message 472)
Posted 29 Jul 2012 by rhb Post: I have now had two C112's that ran 27,000 + seconds; presumably 30 total processor-hours. Other C112's have been 15k - 18k seconds. I did notice at one point that an excessive number of threads were activated during processing. I saw a single yafu thread, 4 shell threads, and 27 gnfs-lasieve threads active at one time. |
|
4)
Message boards :
Number crunching :
Lost a task -- related to Internet going down
(Message 470)
Posted 28 Jul 2012 by rhb Post: I lost task yafu_C111_1343367317_16_0, too many restarts due to lack of heartbeat. http://yafu.dyndns.org/yafu/result.php?resultid=807093 The internet went down at the time it happened. I have had similar problems, though it usually leads to an error 11 (SIGSEGV). If you have any ideas why boinc would restart the task and the heartbeat would continue to not be present, I might be able to debug this (by deliberately disconnecting the internet, if necessary). The router is usually still working fine when I have problems, but the cable modem loses its upstream connection. In some cases, neither BoincTasks nor the Boinc manager can contact the client when it happens. I suspect the OS still thinks the internet is up, and boinc gets in trouble trying to use DNS or something else that times out. I can bring this up on Boinc's forum, but thought I would ask your opinion first. |
|
5)
Message boards :
News :
C112 reached
(Message 463)
Posted 25 Jul 2012 by rhb Post: Yay, go for it. I did a C112 in about 4 hours elapsed time, so I assume that's 16 hours total cpu time. I would probably prefer limiting yafu to 2 or 3 cores, but I'm looking forward to doing some longer ones in any case. |
|
6)
Message boards :
Number crunching :
How much crunching left?
(Message 443)
Posted 11 Jul 2012 by rhb Post: I now have three "relatively" long run-time C110s. All I know is the elapsed time, apparently because Boinc (my version at least) doesn't capture the cpu time of the multiple threads. I had two long-running C110's with the 130.2 application. They ran 18k and 20k seconds, respectively which I figure is 16 to 20 cpu hours. My new (130.3) tasks only ran 11k and 12k seconds. If these are typical, there may be a huge improvement in performance. Of course you would need more examples to be sure, but it looks like the new ones are quite tolerable -- maybe you can push it a little farther? PS. Pardon my sentence structure in the prior post. I noticed they don't parse well, and I apparently can't edit posts. |
|
7)
Message boards :
Number crunching :
How much crunching left?
(Message 438)
Posted 11 Jul 2012 by rhb Post: I got a couple of C110's that ran about 20 hours cpu time. I wish we could see the actual cpu time, I guess it's mostly a Boinc issue not handling multi-thread tasks too well yet. Also, I have 6.58(?) because it's the last Ubuntu version that works, so maybe the new ones version 7 does better. I would be interested in hearing what response you're getting from Boinc about how to see accurate cpu time. I am manually feeding tasks one at a time, which I think is generally a good procedure for an alpha project. I just got new ones, apparently using a corrected 130.3 app and with deadline 2 days with 3+ days cpu estimated. These run immediately when downloaded in high priority. Presumably the long ones will need a full day on my 4 cores if they run as estimated, and will stay active continuously unless I push them out for a while on purpose. I just got the third of these, it turned out the first two ran quickly. This could turn out to be a nuisance unless I open up the spigot. I suspect it will never try to run more than one, and will quit downloading them soon if I let the long-term debt accumulate. Incidentally, I use BoincTasks on a Linux system. |