Page 1 of 1

QuChem prob

Posted: Tue Jun 16, 2020 6:40 pm
by UBT - Mikee
Hi

Having another go at QuChem. Last time they all aborted after a few seconds. They're running now but I have a wee problem!

Despite running at 100% I can only get 7 running at a time. If it was 8 I could understand that it was looking at 8 cores even though I've got 'hyper-threading' on. If I allow another project (say ODLK1) they take up the slack of the other 9 threads.

Don't think there's a problem with memory as I'm showing 7.7GB used, so there's 8.3GB available. CPU usage is currently between 35% and 40% - only got QuChem running now. CPU temp @ 61 degs which is as expected on this PC.

Windows Ryzen 7 1700

One thing I have noticed is that I'm not specifically running VBox as a separate app. Task manager does show 19 instances of 'VBox headless.exe running along with 'Virtualbox global interface' and virtualbox interface' so I assume it's just running in the background but why 19?
In the QuChem library there is a file called 'nwchem_vm_job_0.6.xml' which in edit gives...

Code: Select all

<vbox_job>
  <memory_size_mb>1900</memory_size_mb>
  <os_name>Linux_64</os_name> 
  <enable_shared_directory/>
</vbox_job>
Could it be the 1900 as a size limitation? Straws = clutching! Anyone know before I embarrass myself in the Quchem forum?

Mike

Re: QuChem prob

Posted: Tue Jun 16, 2020 7:07 pm
by ChelseaOilman
In the name of the task does it show t1,t2 or t4? Some of these tasks use more than one thread I believe. You can go into the project preferences to change that. I set my preferences to max 2 threads and got all t1 tasks using a single thread each.

Re: QuChem prob

Posted: Tue Jun 16, 2020 7:34 pm
by UBT - Mikee
Hi

Mine doesn't show max threads - just max CPUs. Everything is showing selected or 'no limit' - 100% resource share. All current tasks are t1 type.

Mike

Re: QuChem prob

Posted: Tue Jun 16, 2020 7:43 pm
by ChelseaOilman
I believe the Max CPU setting is actually threads. If they're t1 though you probably should get a task for each thread. I'm only running 8 tasks on each of my Threadrippers. I'm also running on Linux Mint so no VBox. VBox may have some limitation.

I have Max # jobs 8, Max # CPUs 2 set and I have 8 t1 tasks running on each of my 5 TR boxes.

Re: QuChem prob

Posted: Tue Jun 16, 2020 8:03 pm
by damienh
I think they don't put all jobs in VBOX, and VBOX has been a problem for people in the past. It runs natively in Linux though (no VBOX required!), so that's by far the best way to do it ... but I guess you don't usually run Linux. Sorry, I don't usually use the VBOX version.

Re: QuChem prob

Posted: Tue Jun 16, 2020 8:09 pm
by UBT - Timbo
Hi Mike

I'll put my 2 pennorth in and see if this makes any sense.

In my QuChem project folder, I have the exact same XML file contents as you...so that seems to be "standard" and not dependant on individual hosts or configurations.

I'm guessing that the "<memory_size_mb>1900</memory_size_mb>" setting is how QuChem sets up each VBox instance...so, you have 16 Gb of RAM and each QuChem task is wanting 1.9Gb (even if it doesn't use ALL of what's been set aside for it).

1.9 Gb x 7 instances = 13.3 Gb.

Of course Windows needs some memory too and likewise, you would need some for any GPU tasks you are running...

If you could run 8x QuChem tasks (at 1.9Gb), that would be 15.2Gb, and that leaves little or nothing for Windows or anything else.

As to why QuChem needs 1.9Gb per task, that you need to ask on their messageboards, I suspect.

regards
Tim

Re: QuChem prob

Posted: Tue Jun 16, 2020 8:34 pm
by UBT - Mikee
UBT - Timbo wrote:
Tue Jun 16, 2020 8:09 pm
Hi Mike

I'll put my 2 pennorth in and see if this makes any sense.

In my QuChem project folder, I have the exact same XML file contents as you...so that seems to be "standard" and not dependant on individual hosts or configurations.

I'm guessing that the "<memory_size_mb>1900</memory_size_mb>" setting is how QuChem sets up each VBox instance...so, you have 16 Gb of RAM and each QuChem task is wanting 1.9Gb (even if it doesn't use ALL of what's been set aside for it).

1.9 Gb x 7 instances = 13.3 Gb.

Of course Windows needs some memory too and likewise, you would need some for any GPU tasks you are running...

If you could run 8x QuChem tasks (at 1.9Gb), that would be 15.2Gb, and that leaves little or nothing for Windows or anything else.

As to why QuChem needs 1.9Gb per task, that you need to ask on their messageboards, I suspect.

regards
Tim
Hi

Thanks for the replies. The 7 X 1.9GB theory does make sense - still only using 7.8GB at the moment. May as well run some more ODLK1's in the spare capacity then!

Mike

Re: QuChem prob

Posted: Fri Jun 19, 2020 4:06 pm
by Woodles
Is it just me or has QuChem stopped giving out work to everyone?
Work
Tasks ready to send 60971
19/06/2020 16:03:06 | QuChemPedIA@home | Sending scheduler request: To fetch work.
19/06/2020 16:03:06 | QuChemPedIA@home | Requesting new tasks for CPU
19/06/2020 16:03:07 | QuChemPedIA@home | Scheduler request completed: got 0 new tasks
19/06/2020 16:03:07 | QuChemPedIA@home | No tasks sent
I'm blaming Damien, he seems to be hogging all of them :lol:

Re: QuChem prob

Posted: Fri Jun 19, 2020 4:20 pm
by damienh
Err ... that will be a problem, since I crunch it on zero priority (i.e. with no cache!).

Re: QuChem prob

Posted: Fri Jun 19, 2020 4:21 pm
by damienh
The server status looks OK, so presume that it will recover soon enough ...

Re: QuChem prob

Posted: Fri Jun 19, 2020 4:25 pm
by ChelseaOilman
I feel like this project would be a bad choice for a FB sprint. Another disaster in the making.

Re: QuChem prob

Posted: Fri Jun 19, 2020 4:50 pm
by Woodles
I had it set to zero priority as well and thought that might be the problem, Changing it didn't make any difference.

I think I've been blacklisted, I have work on six hosts but three just gave up.

Edit. I see the problem now, the hosts that don't get work are Windows ones and I seem to remember that QuChem is currently only Linux :oops: