Stop getting Projects or no extra work to store.

[Sandman192]

Describe the bug
This started when I updated one Windows 10 computer to BOINC v7.16.11 (whenever BOINC showed I had an update of BOINC at the time) and another Linux Mint 20 and just started a fresh install of BOINC v7.16.6 two months ago.

Windows 10 side Rosetta and GPU Grid stop getting work with at least 15 other Projects still getting work. I always had a little of each WU for each Project. So I stopped all except Rosetta and GPU Grid. When all the other work finally finished I was getting Rosetta just fine with an extra 10 days of WUs (10 days of work and 10 days of additional work). BUT after a while, I was losing extra work and only running 7 WUs at a time instead of 12 WUs (12 core total CPU) with NO extra WUs. And After a while I'm running back to 12 WUs but still no extra WUs on my computer (just the 12 WUs). As for GPU Grid on the other hand only got 2 WUs (one for each of my 2 GPUs) only with no extra work stored. Doing so when 1 WU finish my GPU was waiting while my computer uploaded and then downloaded 1 WU (with .6 up and 6 down internet speed) (took over a min to start my GPU back up.

Linux Mint 20 all I had is WCG and Rosetta. Started with WCG only. After a while, I added Rosetta. Got some work. I'm guessing due to the extra 10 days of WCG that was on my computer at the time. So I waited till WCG extra WU went down for about 6 days and BOINC would get some more of Rosetta WUs. Both would be 50/50. Nope! Never got any more Rosetta but kept getting WCG. I stopped WCG from getting work and I know after all the WUs finish that Resetta will start getting WU again. But when I start WCG back up BOINC will stop getting work from Rosetta all over again. "I know this because I already tried this on my Windows 10 computer". I had months to test this.

Don't know if this is happing to other Projects. To bussy to find out.

Note: This has never happened till I update BOINC from Windows and fresh install of BOINC on Linux Mint from 2 months ago. Never had this issue before. Been running BOINC on Windows 10 for over 8 years. Linux 2 months.

Steps To Reproduce

1. Still reproducing now.

Expected behavior
A clear and concise description of what you expected to happen.

I always had a little of each WU for each Project (10 days of work and 10 days of additional work). Note: I was told to change the days to whatever they told me to. Nothing happened.

Before this has never happened from 2007 of using older BOINC on Win XP, 7, 8, 10.

Screenshots
If applicable, add screenshots to help explain your problem.

System Information

• OS: Windows10 - Linux Mint 20
• BOINC Version: 7.16.11 - 7.16.6

Additional context
Add any other context about the problem here.

[AenBleidd]

@Sandman192, please attach stdoutdae.old and stdoutdae.txt to analyze details of the behavior you described.
Thanks in advance

[Sandman192]

stdoutdae.txt
Uploading stdoutdae - old.txt…
Had to rename .old to old.txt

[Sandman192]

It's been 12 days of no post. Are you looking into it?

[Sandman192]

It's been 27 days since your last post of Stop getting Projects or no extra work to store. #4110
You ask for "stdoutdae.txt" and I haven't heard from you since.
I had Rosseta only one set to get tasks. BOINC stop getting 10 days of extra tasks downloaded and only had enough for running tasks only. No more. Then after a few days, I had less than the number of tasks than I have cores. Exp: 1 core per task. Well, I have 12 cores (-2 cores for 2 GPUs because of Einstein@home GPU work). So I'm down to running 10 CPU WUs, once per core and after a while, it went down to 6 of Rosseta@home WUs with no extra work and it stayed that way for 2 weeks. I took off Rosseta and had TN-Grid, and WCG only. What did BOINC do??? Only got WCG work!?!?!? And Yes TN-Grid and plenty of work to send. And 10 days of extra work at that! All with 10 days of extra work set.
This didn't happen for years and notice that this problem started after doing the last update of BOINC. Again, I had no problems with this since Windows XP and up to Windows 10.

[Sandman192]

Just to note, Rosseta has no work now.

[Sandman192]

I'm doing a test I'm stopping BOINC from getting all work and waiting for the work I do have is finished. May take a week to finish. No work at all.

For what BOINC did in the past is to grab Work from all Projects, including the extra 10 days of work, and boy I got a lot of work from each Project (21 total) and kept getting all Projects for years. I'll let you know of Projects that have no work to send at all. And I think the new version of BOINC is going to grab all work at first but only going to narrow down to ONE Project ONLY when days go by and never get any from the others. This may take 2 weeks to find out if I'm right.

Windows 10, BOINC 7.16.11.

[Sandman192]

After running out of work for GPU. Was expected. I decided to get more work for 2 GPUs. BOINC got work but only 1 GPU was working. So I got work from another project and still only 1 GPU working. So I got Prime Grid and got the second GPU to work.
But as soon as I restarted my computer from what I got from the last 2 projects that wouldn't run on the second GPU, started running after the restart. 😮?

[Sandman192]

More problems.

As I said before. I had it set to get no more work for ALL projects. As soon as I was down to just 1 WU it still has at least 9 days left to finish. The problem is that it is due on the 14th of Jan and here it is the 25th. 1. why did it get work that couldn't pass the deadline and 2. it should have canceled it when the deadline is up. It's still working on it to the point when I get more work it's still running the last WU and running 12 more WUs totaling 13 WUs on a 12 thread CPU. First time I have seen that.

With just one project left I decided to get more work from LHC "CPU only" and Einstien@Home "GPU only". Well, from Einstien@home for GPU work BOINC is only using 1 GPU only out of 2. But if I got some work from PrimeGrid, Einstien will start working on the second GPU along with PrimeGrid. Again what 😦?

[AenBleidd]

1. why did it get work that couldn't pass the deadline

• because there could be an error of prediction of calculated time and actual task tun-time will be less than predicted
• because not all tasks have constant runtime
• because it could be an issue with task itself (e.g. it's stuck and progress doesn't move)

it should have canceled it when the deadline is up

No it should not. Usually servers have some grace period during which task still could be completed and verified before it will be sent to another user. And even if the last was reported late (it still could be rewarded). You can read a discussion about it here: #2957

running 12 more WUs totaling 13 WUs on a 12 thread CPU

Are you sure all 13 tasks are CPU tasks (and among them there is no GPU tasks)?

[Sandman192]

Are you sure all 13 tasks are CPU tasks (and among them there is no GPU tasks)?

Yes, 13 tasks and 1 GPU task are running. Stayed on 1 GPU until I got some other GPU work from a different Project. Note: On the GPU task that it's taking 1 core each for 2 GPUs. Seems 0.9 CPUs for GPUs share is high enough to have BOINC run only 10 CPU Tasks and 2 GPU Tasks = 12, plus 1 prime = 13 Tasks.
12 cores, 2 GPUs.

[AenBleidd]

10 LHC@home + 1 PrimeGrid = 11 pure CPU tasks
2 GPU tasks share 1 CPU since none of them occupy the full CPU core. So everything here is working as expected

[Sandman192]

Sorry but.

I had 10 CPU Tasks and 2 GPU Tasks before totaling 12 WUs when both 2 GPUs share is 0.9 to 1.0. I get that a lot.

[Sandman192]

Just to note, Do you think when that Prime is done that BOINC will fill that spot? With the SAME type of WUs from both LHC and Einstein. As you can see I'm running 13 WUs right now.

As soon as Prim is done it will go back to 12 WUs as normal 10 CPU and 2 GPU with 0.9 share. Back to 12 Tasks.
I give you a pic when Prime is done to prove my point.

Note: if both of the 2 GPUs share is below 0.9 I'll get 14 Tasks total and if I get only 1 GPU share below 0.9 I'll get 13 Tasks.

[AenBleidd]

10 CPU tasks * 1.0 Core Usage = 10 Core used (of 12)
2 GPU task with 0.9 CPU usage = 1.8 Core Usage
Total: 11.8 Core usage (of 12)

From human point of view I understand your concern.
But BOINC tries to utilize 100% of resources so it run one more CPU task to fill this 0.2 CPU core usage.

And this is a designed behavior right now. And we have an open ticket #2949 to redesign it

[cminnoy]

Unfortunately the BOINC scheduler doesn't try to utilise 100% of your machine. It is well possible that your CPU cores are completely filled, but it won't run any GPU task, even they are ready to start and all other constraints like memory and diskspace are ok.
Also I noticed that if a timeslot is given to a certain project, and that project provides no tasks for your system, BOINC will sit idle and won't try the other projects first.
It is my opinion that there is much room for improvement in the scheduler space.
Another example are tasks that use VirtualBox like cosmology@home. They claim a certain amount of cores (for example 8), but then you see that the CPU is only running on 50% or so (so there is room for more calculations) and no GPU tasks are spawn because the scheduler thinks there is no room anymore to put more.