@@ -57,8 +57,8 @@ The following actions are secured by MFA in JuDoor and therefore a valid token i
* Joining a project
* Adding an SSH Key
* Removing an SSH Key
* Changing someone’s access in a project if you are a PI or PA
Actions that require TOTP are marked with a fingerprint icon.
* Changing someone's access in a project if you are a PI or PA
*Actions that require TOTP are marked with a fingerprint icon.
Currently, MFA is an opt-in feature (can be activated on demand). However, at some point in the future, it will become an opt-out feature (activated by default but can be deactivated if desired).
@@ -84,4 +84,4 @@ We will explain how to run the program in a later chapter.
More Python packages are available as modules.
For scientific computing, the `SciPy-Stack` collection is especially interesting.
See also Supercomputing Environment Template using Python Virtual Environments[venv](https://gitlab.jsc.fz-juelich.de/kesselheim1/sc_venv_template/-/tree/master) to create project-based virtual environments that leverage the already installed Python modules on our systems, which are often significantly more performant than defaults from package managers.
See also [Supercomputing Environment Template using Python Virtual Environments](https://gitlab.jsc.fz-juelich.de/kesselheim1/sc_venv_template/-/tree/master) to create project-based virtual environments that leverage the already installed Python modules on our systems, which are often significantly more performant than defaults from package managers.
@@ -47,7 +47,7 @@ In case you are working on different compute budgets we recommend to set the bud
Every user account on the systems has a home directory (reachable through the `HOME` environment variable) where the user can store his personal files.
However, there is a limit on the volume of data and also the number of files that can be stored in this directory (see more details [here](./budgeting.md#data-quotas)).
Files stored here are also accessible only to you, which can cause issues if collaborators need them and you do not have access.
Files stored here are also accessible only to you, which can cause issues if collaborators need them and only you have access.
Furthermore, the file system performance in `HOME` is reduced.
It is recommended to use `HOME` only for configuration files.
More storage space is granted to computing time projects.
@@ -34,7 +34,7 @@ Two modes of operation are possible:
The `srun` command is used to execute commands on a set of allocated resources.
If no resources are currently allocated, `srun` can infer from its command line arguments what resources are needed, request them from the resource manager and defer the execution of the associated commands until the resources are available.
After the associated commands have been run, the resources are relinquished and running further commands will have to ask for resources again.
After the associated commands have been run, the resources are relinquished/freed and to run further commands you will have to ask for resources again.
This one-shot mode can be useful when you want to interactively run a few quick jobs with varying sets of resources allocated for them.
Run the `hostname` command to see how `srun` will run commands on different nodes than the log in nodes. The `hostname` command lets you see or change the name of your computer (e.g. name of the login or compute node), which is useful for recognising it on a network or setting it up for different tasks. On JURECA and JUSUF, use this command):
...
...
@@ -54,7 +54,7 @@ jrc0454
```
For the JUWELS Cluster and JUWELS Booster, there are a few differences:
The name of the reservation on JUWELS Cluster is hands-on-cluster-YYYYMMDD and hands-on-booster-YYYYMMDD on JUWELS Booster.
The name of the reservation on JUWELS Cluster is `hands-on-cluster-YYYYMMDD` and `hands-on-booster-YYYYMMDD` on JUWELS Booster.
To submit to JUWELS Cluster, you want to be logged in to the Cluster login nodes:
...
...
@@ -92,7 +92,7 @@ Above we have seen four `srun` options:
:::info
The training account budget can be used till the end of the month. After one would need to specify another budget from an active compute time project.
The training account budget can be used till the end of the month. After this, one would need to specify another budget from an active compute time project.
:::
...
...
@@ -541,14 +541,14 @@ LLview is an excellent tool that provides an overview of currently running and f
Your jobs crash and you do not know why?
This is the first place to check.
There is a website for each of our large systems.
You can find the link for every system at the lower left corner of the [documentation webpage of LLview](https://apps.fz-juelich.de/jsc/llview/docu/).
You can find the link for every system in the lower left corner of the [documentation webpage of LLview](https://apps.fz-juelich.de/jsc/llview/docu/).
To begin check out your system of interest, ideally one you have run jobs on.
### Currently Active Jobs
When opening LLview by default it will first show you the list of your currently active jobs, either pending or running. If you are the Principal Investigator (PI) or Project Administrator (PA) for a project and you are in the project view you will see all active jobs from your project. Project mentors also have access to this view.
You filter the list based on the filters below any of the column headings. Clicking on any column heading will cause the jobs to be sorted in ascending or descending order, an arrow will appear next to the column title indicating either ascending (upwards-pointing arrow) or decending (downwards-pointing arrow). In the case a of sort conflict the submission time of a job is used to resolve the conflict, with jobs that were submitted more recently appearing above jobs that started earlier, if sorted ascending. By default jobs are sorted according to ascending job start submit time with your most recently submitted jobs at the top.
You filter the list based on the filters below any of the column headings. Clicking on any column heading will cause the jobs to be sorted in ascending or descending order, an arrow will appear next to the column title indicating either ascending (upwards-pointing arrow) or descending (downwards-pointing arrow). In the case a of sort conflict the submission time of a job is used to resolve the conflict, with jobs that were submitted more recently appearing above jobs that started earlier, if sorted ascending. By default, jobs are sorted according to ascending job start submit time with your most recently submitted jobs at the top.
Values in red in the list indicate that something may be wrong. For example the average load on a node may be high. Note that this is reported in fractions of the utilization of a single core. A 1.0 therefore means that a single core was fully utilized. This value should ideally be as close as possible to the number of cores a node has.
