Latest news

Live update of the Cray Operating System (COS) to release 3.1 on Eiger. The process is transparent to users as no downtime is required: compute nodes have been reserved in groups to proceed with the COS update. Some system libraries path might change due to the COS update: please don't hesitate to contact us if you have any question. The update will improve the stability of compute nodes and will fix the functionality of the MPI hook of Sarus

Eiger is back in service. Please note the following changes impacting the user environment:

- The Cray Operating System (COS) is updated to release 3.0, the Cray Programming Environment (CPE) is now by default 23.12, and the Slurm workload manager is 23.02. Please note that supported applications provided with the default CPE have been re-compiled: we also advise you to rebuild your applications with the new programming environment

- Some applications are no longer provided, e.g. Amber, matplotlib, Scalasca, VisIt. Please check the list of available applications under the corresponding section of the “Alps (Eiger) User Guide” at https://confluence.cscs.ch/x/_gD0E

- Due to incompatibilities with some JupyterLab extensions, JupyterHub can (currently) only launch notebooks, not the JupyterLab interface

- The MPI hook of the Sarus container engine is not fully functional, a fix will be deployed on the system soon

- The number of cpus per task specified for salloc or sbatch is not automatically inherited by srun with Slurm 23.02 and, if desired, must be requested again, either by specifying --cpus-per-task when calling srun, or by setting the SRUN_CPUS_PER_TASK environment variable. See the template batch script provided at https://confluence.cscs.ch/x/XQBYLw

 Eiger maintenance from Monday, June 10 at 8:00 CEST, until Thursday, June 13 (EOB)

The upgrade of Alps is progressing well and we have the opportunity to reduce the downtime by adjusting the upgrade schedule and returning the system today.

Please note that the user environment has not changed:
- We have completed the upgrade of the CSM (the management system of Alps) and the Slingshot interconnect.
- We will require another (shorter) downtime in order to complete the update of the Cray Operating System (COS), Cray Programming Environment (CPE) and Slurm workload manager: it will likely be scheduled during the second half of May.

Please do not hesitate to report any unexpected system behaviour, and note that access to Piz Daint for Eiger projects will remain valid for the rest of this quarter.

The Alps infrastructure is undergoing a major upgrade as we ready the infrastructure towards its full deployment. The system will be upgraded from Monday, 18 March 2024, at 08:00 CET to Monday, 6 May 2024, with no access to compute nodes, login nodes or the scratch file system for the duration of the upgrade. Therefore, please make sure to transfer any important data from the capstor/scratch file system on Eiger before the start of the maintenance using the Data Transfer service.

Eiger will be returned to service with an updated HPE Slingshot, Cray Operating System (COS), Cray Programming Environment (CPE), and Slurm workload manager: as a consequence, you will need to recompile your applications. Due to the extent of this downtime, users will be granted access to the multicore partition of Piz Daint during the upgrade. We apologise for the inconvenience that this may cause

The system will undergo a live intervention on Wednesday, February 28 at 09:00 CET. We expect no disruption in accessing the system or to running jobs. Following the intervention, the buildah and stackinator tools will be unavailable until further notice. Please do not hesitate to contact us if your workflow is adversely affected after the intervention.

The system will be under maintenance from 08:00 CET on Wednesday 24 January 2024 until the end of business on Thursday 25 January 2024, in order to update the scratch file system capstor: a reservation will prevent batch jobs from running during the maintenance window and access to login nodes will be suspended.

• Wednesday, 11 October: /capstor/scratch/cscs (“new scratch”) will be mounted on all Eiger compute nodes (CNs) in parallel to the current /scratch/e1000 , so users can start accessing it. The $SCRATCH environment variable will still be pointing to the old folder /scratch/e1000/$USER
• Wednesday, 25 October (site-wide maintenance): The $SCRATCH environment variable will be modified to point to the new scratch folder /capstor/scratch/cscs/$USER on all Eiger user access nodes (UAN) and CNs. The /scratch/e1000 (“old scratch”) will switch to read-only on all Eiger nodes, therefore you won’t be able to use it in your jobs any longer. Please make sure that all files required by your simulations will be available in the new scratch from this point in time if using the $SCRATCH variable and update your workflow if using the old /scratch/e1000 hard coded in your scripts, otherwise your jobs will fail
  
• Wednesday, 8 November: The old scratch will be unmounted from all Eiger UAN and CN and data deleted permanently

The new scratch will have a soft quota on both disk occupancy and inodes (files and folders), with a grace period to allow data transfer: we will start with soft quotas of 150 TB and 1 million inodes respectively, with a grace time of two weeks. These parameters might be adjusted in the future, if needed

The system will be updated with the following changes impacting the user environment:

- No modules will be loaded by default at login and only the modules cray/21.12  (default) and cray/22.05  will be available. You will need to load the module cray  first, then you will be able to load the modules currently available in the default Cray programming environment (21.12). Therefore, please add the command module load cray  to your scripts and workflows after the intervention. This change is required to prepare the new user environment presented at the User Lab Day 2022 and drafted in the Knowledge Base article uenv user environments. Except for the additional command module load cray , this change should be transparent to you

- The hostnames of Eiger login nodes will be renamed as eiger-ln001 , eiger-ln002 , etc... Please adapt your scripts accordingly if needed

- A fix of the Sarus container engine will be deployed on the system, bringing Sarus back to full functionality

Note that the modules of the Cray programming environment won’t change. We will send additional communications to the mailing list as soon as the date of the intervention will be decided

• Sarus
  • For the PID namespace issue when running multiple ranks per node when using the MPI hook (–mpi),  a workaround is now available by setting MPICH_NOLOCAL=1. (Refer to Sarus for more information)
• Buildah
  • Added Eiger specific documentation (Refer to Buildah for more information)

• Introduction of Slurm policies (low partition and options --account=<project>  or -A <project>  and --constraint=mc  or -C mc)
• CPE update
  • CPE 21.04 will remain the default PE
  • CPE 21.05 is added
  • CPE 21.06 is added (provides gcc/10.3.0 as non-default)
• Container solutions including Sarus, Singularity and Buildah. 
  • Please refer to the documentation below for Eiger's specific Sarus information.