LHC@home

At the moment your server status page shows you've run out of Theory, CMS, and Sixtrack. But it shows 2000 Atlas available. My computers (set to run any type of task) won't get any tasks, the server says:

1545 LHC@home 19-05-2020 16:42 Requesting new tasks for CPU
1546 LHC@home 19-05-2020 16:42 Scheduler request completed: got 0 new tasks
1547 LHC@home 19-05-2020 16:42 No tasks sent
1548 LHC@home 19-05-2020 16:42 Tasks for AMD/ATI GPU are available, but your preferences are set to not accept them

So I changed my preferences to only take Atlas, still no tasks available. What's going on? For the last few weeks I've only been given Theory.

Now, after fiddling with the settings several times, I got a few Atlases, but I can't work out what I was doing to cause it. Can somebody please explain why my computers usually refuse to download Atlas?

---

So I changed my preferences to only take Atlas, still no tasks available. What's going on? For the last few weeks I've only been given Theory.

Now, after fiddling with the settings several times, I got a few Atlases, but I can't work out what I was doing to cause it. Can somebody please explain why my computers usually refuse to download Atlas?

You have to knock three times and ask for Joe.
He is the one with the mustache.

So I changed my preferences to only take Atlas, still no tasks available. What's going on? For the last few weeks I've only been given Theory.

Now, after fiddling with the settings several times, I got a few Atlases, but I can't work out what I was doing to cause it. Can somebody please explain why my computers usually refuse to download Atlas?

You have to knock three times and ask for Joe.
He is the one with the mustache.

That's quite close to the truth. Getting Boinc to do anything remotely like what I'm asking for is almost impossible.

Add to that I have a bit of an overheating problem (I think - strange random problems). I've (stupidly) put most of my computers into the conservatory to make it quieter in here. But when the sun comes out....

Now transporting 5 computers (most without cases, just on shelves of a huge bookcase) to the cooler garage. Tedium. If only there were 4 strong people around I could ask to carry it in one go without stuff falling off.

---

Seriously, does anyone know what's going on here? I left three machines all night on Atlas only, and get other tasks only if those weren't available, yet they have nothing and have gone onto other projects.

---

Seriously, does anyone know what's going on here?

I sometimes see various explanations, but no solutions.
The best fix is to set Rosetta as a zero-resource share backup, so that when there are no ATLAS tasks available, you can do that.

Seriously, does anyone know what's going on here?

I sometimes see various explanations, but no solutions.
The best fix is to set Rosetta as a zero-resource share backup, so that when there are no ATLAS tasks available, you can do that.

I have all my machines doing Rosetta, LHC, and Universe, so running out of work isn't a problem. I was just wondering why the server status claims a full 2000 queue of Atlas, yet I can't get any. I thought perhaps my settings were wrong. I have no problem getting Theory tasks. I've currently got it set to send Atlas, and if none left, send Theory. I just get Theory, as though the Atlas queue was empty, but the server status page claims it's full.

---

There are some obscure (to me at any rate) explanations of how they share the server between multiple projects, and sometimes it is busy with one to the exclusion of another. But I don't think that explains all of what were are seeing. It may be that it is just not a high enough priority for them to fix.

There are some obscure (to me at any rate) explanations of how they share the server between multiple projects, and sometimes it is busy with one to the exclusion of another. But I don't think that explains all of what were are seeing. It may be that it is just not a high enough priority for them to fix.

So too busy handing out Theory to hand out Atlas, or the disk Atlas is on is overloaded so I can't get them right now? Ok, no problem. I just wanted to see some Atlases running because I never seem to get any, and I just upgraded the RAM in two machines so Atlas could run. I've set my account to hand out anything except Theory, but allowed other applications if those aren't available. So far, I'm still getting Theory only. I guess Theory is the most important right now. Well I guess the only way is to deplete the Theory queue so I can get my Atlases :-) [Scuttles off to Ebay to buy more equipment] Think my electric company would give me free power as a charitable donation to science?

---

I guess Theory is the most important right now.

No.
The server doesn't know a priority flag.
If a task is present in the RTS queue and meets all requirements set by the volunteer's request/client limits it will be send out.
Even if we suspect that 1 (out of currently 2) project servers is somehow misconfigured, the other server regularly sends out tasks.
Hence it is a 50 % chance to get a request answered by the "good" server.

Since lots of other volunteers get ATLAS tasks it's more likely an incorrect parameter constellation.
Be so kind as to post
- your web preferences
- your BOINC client limits (RAM, CPU, disk)
- your free disk size
- your app_config.xml (if in use)

I guess Theory is the most important right now.

No.
The server doesn't know a priority flag.
If a task is present in the RTS queue and meets all requirements set by the volunteer's request/client limits it will be send out.
Even if we suspect that 1 (out of currently 2) project servers is somehow misconfigured, the other server regularly sends out tasks.
Hence it is a 50 % chance to get a request answered by the "good" server.

Since lots of other volunteers get ATLAS tasks it's more likely an incorrect parameter constellation.
Be so kind as to post
- your web preferences
- your BOINC client limits (RAM, CPU, disk)
- your free disk size
- your app_config.xml (if in use)

I have three machines which are set to take anything but Theory (I set it this way to find out why Atlas never appeared), then other apps if none available, yet they only get Theory (and sometimes the rare Sixtracks). Here are the requested settings for "xeon1" (if you can see the name from that end), set up only weeks ago. It's the simplest to sort out as it doesn't have any GPUs to confuse matters, and is running nothing but Boinc. It's this computer: https://lhcathome.cern.ch/lhcathome/show_host_detail.php?hostid=10651136 - it's running LHC on weight 3, Rosetta on weight 9, and Universe on weight 1. No other projects are attached. It has 24 cores all in use (2 of 6 core + HT Xeon X5650) and no GPU (just tiny onboard graphics for the monitor). It doesn't do anything but Boinc, it's just got Windows 10 Pro with all the updates, AVG, the latest Boinc, latest Virtualbox and latest extensions, and Remote Utilities Viewer so I can see it from here.

The computer (and the other 2 that refuse Atlas) are in the "default" location. I also have "work" location for 3 other machines which are set to sixtrack only as they're old. These are the web settings for the default location:

Resource share 3
Use CPU ticked
Use AMD GPU ticked
Run test applications? ticked
Is it OK for LHC@home and your team (if any) to email you? ticked
Should LHC@home show your computers on its web site? ticked
General terms-of-use for this BOINC project. ticked
Do you consent to exporting your data to BOINC statistics aggregation Web sites? ticked
Default computer location ---
Run only the selected applications:
SixTrack: yes
sixtracktest: yes
CMS Simulation: yes
Theory Simulation: no
ATLAS Simulation: yes
If no work for selected applications is available, accept work from other applications? ticked
Run native if available? unticked
Max # jobs no limit
Max # CPUs no limit

The Boinc client limits:

Disk:
unticked Use no more than blank GB
ticked Leave at least 5 GB free
ticked Use no more than 90 % of total

Memory:
When computer is in use, use at most 80 %
When computer is not in use, use at most 80 %
unticked Leave non-GPU tasks in memory when suspended
Page/swap file: use at most 60 %

CPU:
Use at most 100 % of the CPUs
Use at most 100 % of CPU time

Other:
Store at least 0.13 days of work
Store up to an additional 0.13 days of work
Switch between tasks every 1000000 minutes
Request tasks to checkpoint at most every 60 seconds

It has a 1TB (930GB reported in Windows) disk of which 879GB is free.
It has 20GB of RAM which is currently 68% used by 6 Theory WUs and 18 Rosetta WUs, occupying all 24 CPU cores.

No app_config.xml in the LHC project folder.

---

If no work for selected applications is available, accept work from other applications? ticked

Untick this one.

Max # CPUs no limit

Set this to 4 to run ATLAS tasks with 4 threads and 6600MB of RAM.
With your setting BOINC will reserve too much memory for an ATLAS-task.

If no work for selected applications is available, accept work from other applications? ticked

Untick this one.

I've done that just now, but I would prefer it to do Theory if Atlas wasn't available. This setting shouldn't have any effect unless Atlas has run out, which I never see on the server status.

Max # CPUs no limit

Set this to 4 to run ATLAS tasks with 4 threads and 6600MB of RAM.
With your setting BOINC will reserve too much memory for an ATLAS-task.

I don't want to limit LHC to only 4 CPUs TOTAL - or is this PER TASK?, but I've set it as you said just now to see what happens. I had assumed that Atlas only goes up to 8 cores, so it will be checking for 10GB of RAM, which is ok on my 20GB system. Or is it trying to reserve memory for 24 cores even though Atlas doesn't do that many?

UPDATE ONE:
Ok, success for now. I changed one of the machines to your suggestions of 4 cores and don't get other apps. And I suspended the other projects on it. It downloaded 8 4-core Atlas tasks and proceeded to run two of them alongside some theories. So I guess 4 cores means per task. Now fiddling with it and another machine to narrow down the problem.

UPDATE TWO:
On another computer (almost identical except it has a GPU, but the same 24 cores and 20GB RAM) I tried changing only one of your suggestions - not to get other apps if Atlas isn't available. It got Atlas immediately, with no core limit set. It downloaded several 8 core tasks.

I conclude that the server is for some reason reluctant to give me Atlas if I will accept something else. Perhaps the scheduler is thinking Atlas is empty when it isn't?

---

I conclude that the server is for some reason reluctant to give me Atlas if I will accept something else. Perhaps the scheduler is thinking Atlas is empty when it isn't?

Not all 'Unsent' tasks are in the send-queue in memory. It seems this send-queue is much smaller and often seems to hold only Sixtracks and maybe some Theory's.
When there are only Sixtracks in it, you get Theory not available, although enough 'Unsent' tasks are shown on the server status page.
The same when you want ATLAS and there are no ATLAS in this memory queue. You get No ATLAS available and maybe get Theory instead when you have 'Send other apps' ticked.

I conclude that the server is for some reason reluctant to give me Atlas if I will accept something else. Perhaps the scheduler is thinking Atlas is empty when it isn't?

Not all 'Unsent' tasks are in the send-queue in memory. It seems this send-queue is much smaller and often seems to hold only Sixtracks and maybe some Theory's.
When there are only Sixtracks in it, you get Theory not available, although enough 'Unsent' tasks are shown on the server status page.
The same when you want ATLAS and there are no ATLAS in this memory queue. You get No ATLAS available and maybe get Theory instead when you have 'Send other apps' ticked.

Thanks, that makes sense. I guess I just have to insist on Atlas and Atlas only if I want Atlas :-)

Just ordered 32GB more RAM to share between the two computers, taking them up to 36GB each instead of 20GB (they'd take 128GB each if I wanted to throw more cash at them). It'll be interesting to see if I can get them both to run three 8 core Atlases at once. It should fit - 3 x 10GB Atlas WUs + 6GB for the OS.

---