Message boards : Number crunching : Please fix the deadlines!
Message board moderation
Author | Message |
---|---|
Send message Joined: 9 Oct 20 Posts: 690 Credit: 4,391,754 RAC: 6,918 |
I'll try one more time. For crying out loud, someone tell whoever is responsible to change the deadlines to something realistic. If you say 1 year, that's what Boinc sticks to. It keeps stopping the tasks to run other projects. I've been told in here you want the tasks back ASAP, not in a year. Although nobody's said what ASAP is. 1 month? If you're unwilling to do this, can you tell me how to manually change the deadline on the tasks I've got at this end? Currently I have to keep suspending other projects to force it to get on with climate prediction, it's getting out of hand. |
Send message Joined: 1 Sep 04 Posts: 161 Credit: 81,522,141 RAC: 1,164 |
Peter - The deadline issue has been discussed repeatedly over the tears. Basically, if the CPDN is set short (say 6 weeks), CPDN can take over your computer. Try changing the project priority for CPDN to 100. Set the project priority for all other projects to 1. |
Send message Joined: 18 Nov 18 Posts: 21 Credit: 6,635,794 RAC: 2,524 |
Peter - The problem with that is that CPDN then takes over and uses all of my cpu cores just for it where I'd rather split time with other projects. Now if you have a cc_config.xml file to share that limits the number of tasks in progress I can do the 100 setting and it will work just fine. |
Send message Joined: 9 Oct 20 Posts: 690 Credit: 4,391,754 RAC: 6,918 |
Peter -I'd prefer it did take over. I've been told by some admins on these forums that they need the tasks back a lot quicker than a year. But nobody has said how quickly. If I leave it as it is, CPDN tasks will be returned in about 2 months. I assume that's not early enough. I have two additional problems that all together make a mess. One is Primegrid always sends out about 4 times as much work as I request. The other is that Primegrid is 24 core WUs on my 24 core machine, and CPDN is 18 1 core WUs. Boinc can't cope with that. How do you fit any combination of 18 of 1 and 6 of 24 into 24 cores? What I've done is something suggested over on the Boinc forums, I edited the client_state.xml to give the CPDN tasks a 1 day deadline. That's made Boinc do those 18 tasks no matter what. Anything else will fit in around it as best as possible. |
Send message Joined: 9 Oct 20 Posts: 690 Credit: 4,391,754 RAC: 6,918 |
If Boinc only had one "debt" scheduler, when downloading, everything would be fine. But it downloads according to your weightings, then decides which to do according to your weightings. The end result is a mess. You should be able to set CPDN to 1 and another project to one, and it would do half of each, but as I said in my other reply, that doesn't always fit when other projects hand out multicore work. Boinc needs to be told by CPDN to get these tasks done when they actually need to be done. Lying to Boinc and saying CPDN has a year to do them means they get left behind.Peter - |
Send message Joined: 9 Oct 20 Posts: 690 Credit: 4,391,754 RAC: 6,918 |
Peter -Over the tears ROFL! Was that a Freudian slip? If they were given out with sensible deadlines, nothing would take over anything, and they'd be done on time. How about tasks that take about 2 weeks to run, with a 6 week deadline? That will only use a third of a PC, giving other projects fair access. The most sensible thing is to have a deadline a few times longer than the estimated time of completion. Say as a human someone told you to get some work done within the year (which unbeknown to you was needed within the month), and several other people told you to get work done within a week. Which would you end up leaving too late? Withholding information from Boinc isn't going to work. |
Send message Joined: 1 Sep 04 Posts: 161 Credit: 81,522,141 RAC: 1,164 |
Peter - Another option is to limit the number of tasks you want to run for each project of you computer. For example, to limit the number of concurrent tasks for CPDN to 4 - 1. Navigate to /var/lib/boinc-client/projects/climateprediction.net 2. Create a file named app_config.xml 3. Place the following three lines in the file. <app_config> <project_max_concurrent>4</project_max_concurrent> </app_config> 4. Save it. 5. In BOINC Manager - > Options, select Read Config files Now the number of cores you have in excess of 4 will run non-CPDN tasks Edit - You wrote - You should be able to set CPDN to 1 and another project to one, and it would do half of each, I think Bonic also takes into consideration how much credit you have on each project. If you have 1000 credit on project 1 and 1,000,000 credit on project 2, Bonic will try to have project 1 "catch up" to project 2. |
Send message Joined: 11 Dec 19 Posts: 108 Credit: 3,012,142 RAC: 0 |
Peter, I run BOINC on Ubuntu Linux and I use LXD containers to isolate projects and pin them to specific sets of CPUs. I did a small right up about it here: https://www.cpdn.org/forum_thread.php?id=9003 It's not for everyone. I see it as the easiest way to dedicate a single machine with lots of cores to multiple BOINC projects by effectively turning it into a few smaller machines with each one using specific CPUs and if needed limits on RAM and Disk usage. |
Send message Joined: 5 Sep 04 Posts: 7629 Credit: 24,240,330 RAC: 0 |
Peter The project people have made it clear that they're not changing anything. It's also been said several times over the years that this project isn't suited for everyone. It may be that you're one of these. As for ASAP, how about we use ASAYC: As Soon As You Can. Years ago there was a batch that run for about 3/4 of a day, so a week or two would do for them. And the latest batches run for 1-2 months, depending on the computer. So 4-6 months is fine. I was just pointing out earlier, that anyone who expects to be able to take almost a year before starting some tasks, may be in for a rude shock, if the researcher gets impatient after 6 months and shuts down a batch. And there are thousands of Windows computer lurking around, just waiting for work if you don't want to run them. |
Send message Joined: 6 Oct 06 Posts: 204 Credit: 7,608,986 RAC: 0 |
I gave CPDN 150% and other projects 1%. Incorrigible they are in hogging all the resources. So I in frustration I detached from the other projects and became a shark lurking around CPDN. Lurking around is all I am doing while "Waiting for Godot to arrive". O' Well, win some, lose some. |
Send message Joined: 27 Jan 07 Posts: 300 Credit: 3,288,263 RAC: 26,370 |
If you have 1000 credit on project 1 and 1,000,000 credit on project 2, Bonic will try to have project 1 "catch up" to project 2. If what you say is true, then that design is faulty. The resource share should try to equalize the recent average credit (RAC), not total credit. |
Send message Joined: 15 May 09 Posts: 4541 Credit: 19,039,635 RAC: 18,944 |
About 1 month on my laptop, less than a week on my Ryzen. ;) Please do not private message myself or other moderators for help. This limits the number of people who are able to help and deprives others who may benefit from the answer. |
Send message Joined: 1 Sep 04 Posts: 161 Credit: 81,522,141 RAC: 1,164 |
Regarding Task Scheduling - Here is a link that describes BONIC scheduling https://boinc.berkeley.edu/trac/wiki/ClientSched Near the top (#4 under goals) honoring resource share over the long term is mentioned. Under Project Scheduling Priority is mentioned using REC (Recent Processing) and resource share to determine priority. Not sure if this helps anyone or just muddies the situation more. |
Send message Joined: 9 Oct 20 Posts: 690 Credit: 4,391,754 RAC: 6,918 |
Peter -I don't really want to do that. When scarce work from projects like this one appear, I want to allow all the cores to do it. And that wouldn't solve the problem I have. If I'm running 24 core primegrid tasks, and get some CPDN tasks, Boinc doesn't want to run the CPDN tasks because it would have to over or under use the number of cores. If the CPDN tasks have a shorter deadline, it panics Boinc and they do run. Setting them to have a shorter deadline manually is working. I got this info from the Boinc forums to edit the client state file: <report_deadline>1608156032.000000</report_deadline> That's a "Unix timestamp" - the number of seconds since the start of the current Unix epoch, and can be converted via, say, EpochConverter to Wednesday, December 16, 2020 10:00:32 PM Edit -I don't think it does. It seems to work on what you've done over the last month or week, not sure of the timescale. Irrespective of my credits, if I allow work from a project I've not done work for over the last week or so, even if that project has a weight 10 times lower than everything else, it will do that and only that project for a bit to catch up. They refer to it as halflife and it's in a config file somewhere, in fact I changed it at some point to a tenth of its value, so it looks back a lot less time, providing a more immediate change when I change what I want to run. |
Send message Joined: 9 Oct 20 Posts: 690 Credit: 4,391,754 RAC: 6,918 |
PeterI was suggesting a way it could be made suitable for everyone. All that needs to be done is to inform the Boinc clients of a real world deadline, then it can act accordingly. I see no advantage in letting Boinc think it can take a year over it if results returned in a year are worthless. Even if the deadline is too short, if it's started the task, it will still finish it. The WAH2 tasks I'm running just now run for 1 week to 1 month depending on computer speed. 1 week on a new Ryzen 9 and 1 month on a 10 year old Xeon. So you're fine with them coming back in about 3 months? They'll all be back in 1 month anyway, since I boosted the deadline in the client state file :-) I'm surprised scientists are ok with stuff back in a matter of months. Maybe they have other things to do while waiting, like analyzing the stuff they sent out 6 months ago. In my mind I'd get impatient! Can I assume that if I'm receiving credit each Thursday, that the tasks I'm running are still needed? If you don't want them, will a cancel notice be sent out to stop them running? |
Send message Joined: 9 Oct 20 Posts: 690 Credit: 4,391,754 RAC: 6,918 |
Peter,That's not for me, I run 3 projects (LHC, MW, Primegrid) that do multicore tasks. Easier to let Boinc manage all the cores as one unit. |
Send message Joined: 11 Dec 19 Posts: 108 Credit: 3,012,142 RAC: 0 |
I split my Ryzen 9 3950 into 4 systems each with 4 physical cores and if needed the 4 complimenting logical cores from SMT. It works great. |
Send message Joined: 9 Oct 20 Posts: 690 Credit: 4,391,754 RAC: 6,918 |
I split my Ryzen 9 3950 into 4 systems each with 4 physical cores and if needed the 4 complimenting logical cores from SMT. It works great.Wow, that seems like a lot of effort. I can tweak Boinc to make it do pretty much what I need. I usually leave it on automatic, using only project weights to make it do the proportions I want, then shove it in the right direction now and then if it's not doing what I need, like leaving some tasks too late. By the way (and your local prices may vary), I just bought a Ryzen 9 3900XT because it's a lot cheaper and not much slower than your Ryzen 9 3950. Here in the UK: Ryzen 9 3900XT: £374, 33201 CPU Mark. Ryzen 9 3950X (I can't find a 3950 on sale anywhere): £590, 39282 CPU Mark. That's far too much money for a small increase in speed. |
Send message Joined: 31 Aug 04 Posts: 29 Credit: 3,971,150 RAC: 884 |
I was under the impression that Project Admins set the deadline for each batch released if they have a batch with short expected run times, set a short deadline If longer expected run times, a longer deadline Or am I misunderstanding something? I currently have two WUs with expected run times of ~12 hours, and a deadline of 1 year, which is absurd |
Send message Joined: 15 May 09 Posts: 4541 Credit: 19,039,635 RAC: 18,944 |
I was under the impression that Project Admins set the deadline for each batch released I am confused. You only seem to have one computer attached to CPDN and it only shows one task which on an I7 I would expect to take three to five days. CPDN has very few tasks that last less than 3 days even on the fastest computers these days. Please do not private message myself or other moderators for help. This limits the number of people who are able to help and deprives others who may benefit from the answer. |
©2024 cpdn.org