Hydra move to VUB Etterbeek campus completed#

../../../_images/hpc-cables-02.jpg

We are happy to announce that the move of Hydra to the new data center of VUB in Etterbeek campus is complete. The cluster is now back open to VSC users.

Our apologies for the three days of delay with respect to our original plan. We experienced some unforseen difficulties with the network configuration in the new emplacement of the cluster. Not only affecting the new storage, but also other critical systems such as the login nodes.

Nonetheless, this extra time has been very productive and we did solve all outstanding issues with the cluster. This means that we fulfilled all our goals for the move and Hydra is coming back online in full. Including the integration of the new storage for VSC_HOME and VSC_DATA.

The change of storage is a critical change for our users. The old storage of Hydra was nearing its end-of-life and the risk of data loss was very high. We already hit a situation very close to data loss during the power outage in March. Therefore, the renewal of the storage is a much needed improvement that also comes with other advantages:

  • improved reliability: data in VSC_HOME and VSC_DATA will be stored in a filesystem with redundancy and off-site replication. We can also carry out maintenance on the new storage without inflicting any downtime in the cluster.

  • larger capacity: up to 1 PB of storage (in total), which is easily expandable and can be made accessible to our users and virtual organizations upon request.

However, these improvements of the new storage come at the price of a lower I/O performance. While it was acceptable to execute jobs reading or writing data from VSC_DATA in the old storage, now the performance of your jobs will be impacted. The new storage is very performant, but it is not meant to work as scratch. Therefore, all users should revise their workflows to ensure that their jobs read and write data directly from scratch (VSC_SCRATCH or VSC_SCRATCH_VO).

We have update our documentation about HPC Data Storage in Hydra to reflect these changes. We will add examples to automatically handle data copy between VSC_DATA and VSC_SCRATCH in the next days. In the meantime, if you need any help to adapt your job scripts, please contact VUB-HPC Support.

Finally, the 12 Nvidia Kepler GPUs in Hydra have been decommissioned as planned. They are no longer available.

If you have any questions or concerns, please contact VUB-HPC Support.

../../../_images/move-09.jpg

The new house of Hydra in the VUB data center.#