Interactive Tool - multicore? startup time

1

We are running our IT (“PhysiCell Studio”) on the EU instance. It’s a multicore (OpenMP) simulation framework and I’m just wondering if we are actually getting the requested number of cores at runtime. Is there a max number of cores allowed?

Unrelated to that, I’ve been comparing the startup time of this IT (from clicking “Run” to it being ready to “Open”) and it seems to average 1.5-2 mins. On nanoHUB, where we’ve hosted the same tool for a couple of years, it takes about 20 sec. I’m just curious about the differences in the hidden workflows, in case someone could provide more details about the Galaxy IT.

2

Regarding the latter comparison, I asked one of nanoHUB’s sysadmins and learned this, fwiw:

Before the user clicks on “Launch” your tool will be installed and ready within nanoHUB’s /apps/pcstudio directory which has each published version of your tool stored, along with a dev version that you can test before publication. To clarify, the pcstudio tool itself is not Dockerized. However, when tool sessions are launched it is as a Docker container running the operating system needed to run your application. The directory where your tool is stored is mounted to the tool session container and launched using the invoke script stored in the middleware directory in your git repo.

3

I partially answered my first question - I confirmed that I’m getting 4 cores from OpenMP.

4

For EU you should only get 1 core. infrastructure-playbook/files/galaxy/tpv/interactive_tools.yml at master · usegalaxy-eu/infrastructure-playbook · GitHub

The startup time depends on the a few factors:

  • scheduling time: is the cluster full so that you need to wait for the job to be scheduled and running
  • Docker image size: if the job is exectured on a Node which has not yet a copy of pcstudio locally, then the image needs to be pulled - the larger the image to longer it will take
  • Container startup time: how long does the container to be online

That should be all I can think off.

5

Thanks. If I’d heard about the 1 core limit in the past, I’d forgotten. Is that a hard limit or not? I see at least one IT (interactive_tool_ilastik) in that link that has cores > 1.

And I realized shortly after I “confirmed” in a test program that I could get 4 cores that, yes, even a 1-core machine will let you think it has 4 cores in OpenMP.