HPC lesson updates and clarifications from Kathleen

[rkhetani]

• data storage & memory on O2 are in units using base 2, not units using base 10 (e.g. tebibytes not terabytes. TiB = 1024 GiB, TB = 1000 GB). HMS IT had been using the units that people colloquially use- such as terabyte, gigabyte, etc. - but have been technically incorrect with these units. The amount of storage that folks have been using/have access to has not changed with our change in terminology. The distinction is important for the billing aspect, as we charge for compute usage (with a RAM charge for GiB/hour, among other factors) and for storage usage (TiB/year). More details on billing rates here
• for the sentence “There are several compute nodes on O2 available for performing your analysis/work”, do you mean several types of compute nodes? That is true, or you could also say “There are several hundred compute nodes…” which is also true. The sentence as is sounds like it is missing a word.
• Memory request would be in gibibytes, not gigabytes for --mem 1G
• This won’t be relevant for the workshop itself, but if folks are submitting jobs and are in multiple Slurm accounts (e.g. labs/groups), they’ll need to specify an account for an srun or sbatch job to count under with the -A parameter. You can check if you’re in multiple Slurm accounts by running sshare -Uu $USER. More details on -A and Slurm accounts/unix accounts here
• The wiki link for -t is broken, missing a dash, use this: https://harvardmed.atlassian.net/wiki/spaces/O2/pages/1586793632/Using+Slurm+Basic#Time-limits
• Same thing for -c: https://harvardmed.atlassian.net/wiki/spaces/O2/pages/1586793632/Using+Slurm+Basic#How-many-cores?
• And --mem: https://harvardmed.atlassian.net/wiki/spaces/O2/pages/1586793632/Using+Slurm+Basic#Memory-requirements
• And O2 wiki sbatch reference link: https://harvardmed.atlassian.net/wiki/spaces/O2/pages/1586793632/Using+Slurm+Basic#sbatch-options-quick-reference
• sbatch job submission is using 400MiB
• module load can modify additional environment variables than $PATH, specifics are probably not relevant to this workshop, though
• We’re starting to move away from gcc/6.2.0 and are building new tools with gcc/9.2.0, but the majority of modules have been built with gcc/6.2.0
• For the filesystems part, it’d be helpful to link to here, as it has links for requesting group directories (under the Active Compute section). Also, a caveat that off quad folks will have to pay for their group directories. Home, scratch directories are free for everyone. Also, /n/cluster/bin/scratch3_create.sh needs to be run from a login node. The script will give you an error message to this extent if you run it from a compute node, but sometimes folks don’t read :bloblul:

[rkhetani]

Some links may be broken, but I think @Gammerdinger checked that when he went through for ChIP-seq.