Changes from biocluster to marbits

Besides having a cooler name and logo, several things have changed from biocluster to marbits:

Hardware

Older nodes are still online, but we have added 10 supermicro nodes, with newer processors (24 physical cores per node; 48 threads) and more RAM (8 nodes with 256 GB and 2 nodes with 774 GB). They are faster and more capable and will endure most of the jobs. See more system specs here.

Software

All HPC clusters need an environment that allows to build a cluster image, provision nodes, install packages, etc. Biolcuster used Rocks (v6) and marbits uses openHPC (v1.3.4). This change is transparent to you the user.

The job scheduler (the program that manages the queue and allocates resources for your jobs) has changed from Sun (now Oracle) Grid Engine, SGE to Slurm. You’ll notice that the options that your jobs used have changed and that you’ll have to submit jobs with sbatch instead qsub… See here to learn how to submit jobs in marbits.

Programs are installed system-wide and they are loaded in modules (see here). It is very similar to the last biocluster era, when all programs were loaded in your path dynamically.

Storage

We used to have a very fragmented NFS storage that was slow and difficult to use. Now our data are in a lustre filesystem that we share with other groups and departments at CMIMA. See here to know more nad to read about some rules to use it.