gw4-isambard / docs Goto Github PK

This could be a bit more hierarchical with overview linking to different pages for each of the substantially different components.

ProxyJump is better supported.

Add note to system status about upcoming maintenance

Good artists copy; great artists steal.
-- Pablo Picasso (allegedly)

Before the ARCHER service closes and they move on to ARCHER2 which has slurm, it might be good to copy the information from their Troubleshooting Guide as I have used this several times when debugging PBS issues on Isambard.

We refer new users to the list of projects, it would be helpful to list likely projects since the SAFE list is quite long and the project names don't start with the institution name so it's not searchable either.

https://github.com/gw4-isambard/docs/blob/master/user-guide/connecting.rst#L53

Typo is not in the config block so it's not urgent.

User had an OOM issue which occurred due to a difference in inputs for similar jobs causing and increase in the memory requirements from ~220GB to ~290GB.
The jobs ran on low node numbers to weren't directly impacted but if they had been allocated the new nodes nodes with higher memory confusion for the user and diagnosing the issue would have been greater. At very least we should have clearer advice around this instructions for requesting higher-memory nodes in jobscripts:

#PBS -l mem=500GB

If this becomes a recurring issue then we could also consider

User reported issue with current instructions for WinSCP that doesnt cover ssh keys.

Several tickets including 284 have requested installation of libraries, suggest creating how to guide on creating venvs and installing libraries on the different components might be useful.

In some places the documentation quotes the number of ThunderX2 compute nodes as 328 (consistent with a core count of 20,992) and in others it is quoted as 329 (or 160+169 nodes).

Dear @jrprice ,
I’m a user of the ISAMBARD and I’m interested in using a plumed patched version of Gromacs. I can see that Gromacs has been installed in ISAMBARD. I wonder if I can have the installation script so that I can install the plumed patched version of Gromacs? Thank you.
Kind regards,
Zhiyi Wu

This repo uses James Price's Travis CI to build the public facing docs, we should move it to Github Actions within this repo.

Add link/details to use of puttygen if necessary to generate
Add link/details to use of pageant to store keys to ease access via bastion
Add link/details for adding key to putty connection

Also check that advice for Linux includes safe ssh use e.g. point to: https://arc-lessons.github.io/security/00_schedule.html which also contains advice for putty via references as well.

A user reports

On this page:
https://gw4-isambard.github.io/docs/user-guide/MACS.html
Under the bit on the Power 9 nodes, it says:
"..they are representative of the type of node used in large-scale HPC”
This should really say “a type of node” rather than “the type of node”, as the latter implies there are no other types.

As potential user asks

looking at what queue configurations might be available, in particular walltime restrictions [...] but was unable to find these on the Isambard online documentation.

Add note on Power 9 unavailability to https://gw4-isambard.github.io/docs/user-guide/phase1.html

Post update default gcc version has changed
VASP is built with 7.3.0
Module swap includes version so these currently fail as per RT ticket #701
change to module swap gcc gcc/7.3.0 to fix

Add specific contributions guide as CONTRIBUTING.md
Update README.md to reflect this

Following upgrade to partitions need to update information whats available on the system. Also need to clarify what is a phase. Are we renaming this service Isambard 2 or have phases have different meanings?

vtune use does not appear in user docs
as per support ticket 74 re gui not running on login nodes and response from JPrice:

I looked into this today (sorry for taking so long!), but it seems like this missing library is tied to the graphics driver, which we don't have on the login nodes. We do have the required drivers on the P100 compute nodes though, so you can `qsub -X -I -q pascalq` and run `amplxe-gui` from there. Is that sufficient for your needs?

Testing GitHub Slack integration

We found with PPMD that you need to make careful choice of -cc flag to arrange jobs if you are using hybrid mpi+openmp.

Add new page covering options aprun options including some description of thread placement options: `-cc {cpu[default], none, numa-node, depth}.

We have had a user request for more detailed node info

I need to run my code with more memory than is currently available per core per node. To counter this I plan to under-fill each node, but I am uncertain how many NUMA regions, if any, there are per node.

Add documentation for default compiler environment and how to switch between on xci if this is not yet clear.
Add details of recommended optimisation flags and information e.g. about vectorisation.

Some details in rt ticket 83

A user reports

Would it also be possible to get information about budgets and max node/core size per job? As I cannot find this in the GW4 Isambard user guide.

Reminder that we had a user using Ubuntu that used gpg-agent. Seems we might need to modify the arc-lessons/security material.