Login node usage


* Compute Canada documentation: https://docs.computecanada.ca/wiki/Running_jobs

The Guillimin login nodes are the computers that users log into when they first connect (ssh) to guillimin.hpc.mcgill.ca. The names for these nodes have the form lg-1r14-n## or lg-1r17-n## (e.g. lg-1r14-n04). The purpose of these nodes is to provide an interface for users to manage their files and submit jobs to the worker nodes (see: Submitting your job). They are not intended for running CPU or memory intensive jobs, and doing so can negatively affect the work of many other users who share the node.

What tasks should be submitted to the worker nodes?

In general, there are many users simultaneously logged in and working on each login node. These login nodes are not as powerful as the worker nodes. Therefore, the computing resources of the login nodes can be spread thin when many users are using them. Sometimes, a user can start a process which uses so much CPU or memory that it impacts the work of everyone else using the node. If your task meets any of these criteria, you should be submitting it to the batch system instead of using the login node:

  • The task will run for more than 5 minutes. Note: compiling large applications or libraries is an acceptable use of login nodes, even if the complete build takes more than 5 minutes.
  • The task will use more than 4 GB of memory
  • The task requires that several heavy processes be launched simultaneously. Note: if you compile an application or a library, make sure to use a maximum of 1 process
  • The task uses any kind of parallelism (MPI, OpenMP, Pthreads, etc.)
  • Please be aware that some software (such as Matlab) has automatic parallelism that should be deactivated before running on the login nodes. Please check your processes using the 'top' command to ensure that they are not using more than 100% CPU. Check the documentation of your software package for instructions on how to run in single processor mode.

Many tasks can be performed by the worker nodes, even if it may not be obvious how to submit them. If part of your workflow on the log-in nodes violates any of these guidelines, please contact our team for advice on how this process can best be run without disturbing other users: This email address is being protected from spambots. You need JavaScript enabled to view it.

How to do interactive or graphical work on the worker nodes

Many users have resource-intensive jobs which can't be submitted as a script because they require interactivity and/or a GUI interface. For example, interactively creating a plot or movie from a large data set may require a graphical program that consumes large amounts of CPU and memory. These tasks can be done using an interactive session on a worker node. In the example below, RAPID is your RAP id, of the form xyz-123-ab:

$ qsub -I -X -l walltime=1:00:00 -A RAPID
qsub: waiting for job 15746351.gm-1r16-n04.guillimin.clumeq.ca to start
qsub: job 15746351.gm-1r16-n04.guillimin.clumeq.ca ready

Begin PBS Prologue Thu May  1 09:09:36 EDT 2014 1398949776
Job ID:         15746351.gm-1r16-n04.guillimin.clumeq.ca
Username:       user
Group:          user
Nodes:          sw-2r12-n36
End PBS Prologue Thu May  1 09:09:36 EDT 2014 1398949777
[user@sw-2r12-n36 ~]$ 

The '-I' (capital i) option will instruct the scheduler to start an interactive session and give you access to a command line on the worker node. The (optional) '-X' option will forward X windows through the login node so that you can use graphical programs. You may also specify '-l nodes=1:ppn=12' if you require additional processors or memory for your job (see: Submitting your job). Inside this job, you can access the same software and files that you can access from the login nodes. However, you will have dedicated resources for your task so that your work does not disturb other users.

Running services on the login nodes

In special cases, we make exceptions to the guidelines above for users who need to run services (such as license managers) that must run perpetually on the login nodes. If you use the login nodes in this way, please contact our staff for advice, and to be sure that your process does not get terminated: This email address is being protected from spambots. You need JavaScript enabled to view it. .