Compute Grid Tip of the Month - March 2017

March 3, 2017

Choose your resources efficiently

Using the HBS Compute Grid is a great opportunity to scale your research beyond what you can do on a desktop or laptop. But, as a shared resource, we all have a responsibility to use the resources appropriately. The biggest impediment we face is Grid users over-requesting RAM and CPUs: once requested and allocated, these are not returned to the general pool until your job has finished or you have quit your program in the interactive session (the programs from the Applications drop-down menu in NoMachine). So, if you ask for more resources than your job or program needs, those “extra” resources are, in effect, wasted/unavailable to be used by other users. Please help other users out by choosing your resources efficiently when running jobs and program (“Take What You Need, but Need What You Take”).

To understand how much RAM you really need, towards the end of your session, enter the following command in the terminal (Applications > Accessories > gnome-terminal):

bjobs -l | grep -E "Project|MAX"

The number returned will tell you the maximum amount of RAM that your session(s) used. So next time when starting a session, you should pick the RAM size (-5g, -10g, etc) that is closest to your max usage.  

Following this guideline, esp. when starting new work or working with a new dataset, not only allows more people to use the compute grid simultaneously, but also allows you to run more simultaneous sessions within the current 12 CPU and 80 GB RAM resource limits.