Running boinc with multiple projects: best practices?
So I'm relatively new to this multi project thing.
I've been attempting to figure out which CPU projects favor Intel's (usually anyway) advantage in IPC and AVX (Primegrid) and which favors AMD's multithreading (Amicable). In doing so I've found a few oddities that maybe someone can help answer.
I've noticed that Boinc really doesn't know how to manage resources I set for a project. I have Universe set for at 10 total resources and amicable set for 100 (or the default), just as an example, yet Universe has been steadily crunching for the last week or more. I've been suspending Universe in the meantime and micromanaging myself. Why have resource management if Boinc clearly doesn't respect it?
That being said, what are some best practices for running multiple CPU projects? How many days of boinc cached work is a reasonable amount (apart from bunkering). Right now I have 0.1 days minimum with up to an additional 0.5. It seems like it takes boinc a while to settle down and figure out just how much work amounts to 0.5 days, though. It either overcompensates or doesn't fetch enough because it doesn't think the CPU will be able to finish what it has. Maybe I just need to be patient and wait a few months? Or should I be setting this to more or less? Deadlines are being met so maybe I'm just overcomplicating things.
What do folks use to track actual daily credits per machine rather than RAC? That seems to be wildly varying for some odd reason on the same computer with the same project, and we won't even talk about two different projects...I've been using BAM, but the "new" design is very much not screen reader friendly.
Has anyone tried the projects with advantages in Linux inside of a virtual machine? Are the gains worthwhile? Or even the Linux subsystem within Windows 10.
Anything else I should bare in mind?
Thanks!
Re: Running boinc with multiple projects: best practices?
Quote:
Originally Posted by
wolfman1360
So I'm relatively new to this multi project thing.
I've been attempting to figure out which CPU projects favor Intel's (usually anyway) advantage in IPC and AVX (Primegrid) and which favors AMD's multithreading (Amicable). In doing so I've found a few oddities that maybe someone can help answer.
I've noticed that Boinc really doesn't know how to manage resources I set for a project. I have Universe set for at 10 total resources and amicable set for 100 (or the default), just as an example, yet Universe has been steadily crunching for the last week or more. I've been suspending Universe in the meantime and micromanaging myself. Why have resource management if Boinc clearly doesn't respect it?
That being said, what are some best practices for running multiple CPU projects? How many days of boinc cached work is a reasonable amount (apart from bunkering). Right now I have 0.1 days minimum with up to an additional 0.5. It seems like it takes boinc a while to settle down and figure out just how much work amounts to 0.5 days, though. It either overcompensates or doesn't fetch enough because it doesn't think the CPU will be able to finish what it has. Maybe I just need to be patient and wait a few months? Or should I be setting this to more or less? Deadlines are being met so maybe I'm just overcomplicating things.
What do folks use to track actual daily credits per machine rather than RAC? That seems to be wildly varying for some odd reason on the same computer with the same project, and we won't even talk about two different projects...I've been using BAM, but the "new" design is very much not screen reader friendly.
Has anyone tried the projects with advantages in Linux inside of a virtual machine? Are the gains worthwhile? Or even the Linux subsystem within Windows 10.
Anything else I should bare in mind?
Thanks!
Maybe I can answer some of this.....
In general you can run two different projects per machine, one GPU the other CPU. Some projects have both a GPU and CPU application. Generally GPU and CPU does not mix on the same project. If you wish to run two different CPU projects on the same computer you need a resource controller like Process Lasso where you can dedicate a certain number of cores to certain processes.... Then use the app_config file to limit the number of cores a project can use.
Most people run one GPU project and one different CPU project leaving a CPU core open for each instance of a GPU application running.
Cache: Me, I run with the full 10 days and 10 extra days in boinc cause I want the largest cache possible.
Intel/AMD...... Linux/Windows...... Easiest way to find out is to look at the projects host stats page, it will tell you what the host producing the most are running for both CPU & OS. Generally, Most projects favor Linux particularly the grid projects.... the CPU difference isn't that far apart at this point. back in the day, AMD's ruled, then it was Intel's, today on the newer hardware it's kinda even, neither has any real advantage in my opinion, the latest Threadrippers really haven't been out long enough to tell but look like they will be a bit of a step forward for AMD.....
AS far as Boinc and process management, it's slow... it will eventually figure out what is what it considers the optimum cache. Boinc's maximum cache is 1k WU's/per machine. Boinc on it's own will not give you 1k WU's, (and some projects limit the # of allowed WU's) there are workarounds......
For Boinc management purposes most around here use BoincTasks a good software even if it is a bit clunky.... It is network aware also and much easier to use that Boinc Manager... Been crunchin for over 20 years I know very few people that use BAM...... (the first time I looked at it, I ran the other way ----------------------> as far as I could, way to convoluted for me)
Linux VM's for crunchin, plain flat work.... the long and the short of it....
Most of us get our stats from Free DC, the forum we are typing on is hosted on FDC and Bok is good people! (he is also a member here)
I hope that answers some of your questions, I don't know a whole lot, but what I know, I know......
And welcome to the forums, keep askin, there will be someone here with an answer....
Re: Running boinc with multiple projects: best practices?
Let me add a bit more info on BOINCTasks. You run it on a single machine, and you can monitor all your machines combined. Saves you from hopping from machine to machine to see what each is doing. Very slick.
Re: Running boinc with multiple projects: best practices?
Quote:
Originally Posted by
zombie67
Let me add a bit more info on BOINCTasks. You run it on a single machine, and you can monitor all your machines combined. Saves you from hopping from machine to machine to see what each is doing. Very slick.
Boinc Manager is network aware also, and you can run all your networked boinc machines from one machine running Boinc Manager, you just can't have all of your networked machines in one window at the same time like BOINCTasks lets you do.....
That alone makes BoincTasks leaps and bounds better than Manager...
Re: Running boinc with multiple projects: best practices?
Okay...I've been living under a rock for 2 years. Or maybe I tried out BoincTasks and gave up on it since there is a bit of work to get it to play nice with NVDA (screen reader). It likes to refresh and yank my focus back to the top of the list of tasks among other things, like unlabeled edit boxes in settings. Still I'm stubborn and persistent and working on accessibility issues is my job, so I'm slowly learning my way around it. Incredible. I mean, the info is just right there. Everything you could want or need to control the client, tasks, projects, and if I'm reading this right even CPU control using tThrottle. ....goodbye bam. I never used it much anyway and you are so very right, it's very confusing and convoluted.
Tomorrow will be working on all the machines on the network. Can I also use this over the internet, if I'm away from home? I just skimmed the manual and I'm only seeing computers over the network though to be fair I didn't get very in depth.
Process Lasso...boy oh boy. How do you go about divvying up what processes to run when and on what cores? I don't really have a lot of powerful gpus - Rx570 and gtx 1080 are my main two along with a 1050 ti in my x1 extreme, but that then yanks the CPU's TDP to 15 W instead of 45 and I have no clue how to fix that. Oh, and I guess I can count the poor old hd6450 and gtx 670M...but everything else are cpus. Having no vision I'm glad I can use the gpus I do have for some good.
I'll take a look at that other stats site you mentioned tomorrow too. Me and boincTasks are going to become fast friends in the meantime.
I'll have to spin up a Linux vm at some point and play with it. Using Orca via the gui got me nowhere and initial setup through the terminal was a little confusing for me but doing it repeatedly and getting more familiar with Linux commands in general would do me good anyway.
Re: Running boinc with multiple projects: best practices?
ProLasso is actually a pay-to-play program if you want ALL the features. The lifetime license costs something like $50 for 5 computers. If you don't buy the license some features quit working. The ones we primarilycare about, CPU affinity and Priority, continue to work :D
Re: Running boinc with multiple projects: best practices?
Quote:
Originally Posted by
wolfman1360
Okay...I've been living under a rock for 2 years. Or maybe I tried out BoincTasks and gave up on it since there is a bit of work to get it to play nice with NVDA (screen reader). It likes to refresh and yank my focus back to the top of the list of tasks among other things, like unlabeled edit boxes in settings..
This also happens to me on BoincTasks.
When I go to attach or add a project, I can't get my email address and password typed in before the focus gets ripped away from the text box.
I never learned to type but I'm not that slow.
It can sometimes take two to three partial attempts to get the info typed.
