[live updates] VUB HPC clusters moving to Research Park Zellik#
Hydra and Anansi will move to the data center in the Research Park Zellik on the week of September 7th. The clusters will be fully shut down on Monday September 7th at midnight and reopen to users a week later, on Monday September 14th.
During this period, it will not be possible to connect to Hydra or Anansi and jobs will not run. This restriction includes the following services:
SSH connections through the terminal interface
OnDemand sessions through portal.hpc.vub.be
Remote sessions with tunneling such as VS Code applications
Globus collection VSC VUB Hydra
Once the move starts we will publish regular updates on this same news post with details on the progress of the operation. See the Timeline of events and updates below!
Impact on Storage#
We are currently taking backups of the scratch storage in Hydra prior to the move. These special backups will guarantee that there is no data loss even if some hardware is physically damaged in the process.
The storage system for VSC_HOME and VSC_DATA is partially affected by this move. This storage is not moving but needs to be reconfigured. Hence, there will be a disconnection of VSC_HOME and VSC_DATA from Monday September 7th at 10:00 (CEST) until 15:00 (CEST). During this period, users will not be able to login to any VSC cluster on other sites.
Impact on Compute#
Users urgently needing compute time during the move can use the UGent Tier-2 systems as they are similar to ours. Check out the VSC documentation on how to access them, and the HPC-UGent documentation for usage info.
The last 5 days before the move, only jobs that can end before September 7th at 00:00 (CEST) will start. Jobs left in the queue before the move will be kept in the queue and will automatically start once the move is finished and the cluster is back online.
Datacenter in Research Park Zellik#
The change of location from the data center room in building G of Etterbeek campus to a modern data center such as Penta Infra BRU01 will bring improved infrastructure at all levels. Greener and more stable power supply, more efficient cooling and better security. Hydra and Anansi will not share the space of the VSC Tier-1 cluster sofia though, they will be placed in a different room exclusively used by VUB alongside other systems of the university.
Timeline of events#
The timeline below shows the main milestones during this operation and it will be updated regularly. We will also notify of any changes to the planned schedule.
If you have any questions or concerns about the move, please contact VUB-HPC Support. User support will remain fully operational during the move.
Date |
Time |
Action |
Status |
|---|---|---|---|
04/09 |
17:00 |
Complete inventory, labelling and order of new components |
|
07/09 |
00:00 |
Hydra and Anansi clusters shut down |
✓ Completed |
07/09 |
06:00 |
Final data sync of VSC_SCRATCH backup completes (see Impact on Storage) |
✓ Completed |
07/09 |
15:00 |
Login to non-VUB VSC clusters re-established for VUB users (see Impact on Storage) |
|
07/09 |
18:00 |
Hardware arrives at Research Park Zellik (hopefully in one piece) |
|
08/09 |
17:00 |
Storage of VSC_SCRATCH cabled and operational |
|
10/09 |
17:00 |
Compute nodes of Hydra and Anansi cabled and operational |
|
11/09 |
17:00 |
Login to Hydra re-established |
⚙ Pending |
Updated on 04/09/2026
Most preparations for the move of the clusters next Monday are ready. We reviewed our inventory of components, all systems moving to the new data-center are clearly labelled to be easily located and we already received the delivery of almost all new components needed to connect everything in the new location.
We completed the first full backup of VSC_SCRATCH, which will receive a final data sync once the clusters are shut down.
We got meters of new network cables. The network fabric in the data-center of RPZ is different from in G0, so we needed to adapt the connections of our systems.#
Racks are color coded to ease the transition of the nodes to their new destination. The zen5_mpi nodes go to the yellow rack.#
Nodes are labelled front and back. This is the back of some ampere_gpu nodes that go to the blue rack.#
The zen4 nodes go to the green rack.#
Updated on 07/09/2026
15:30 Compute nodes of Anansi and Hydra, the scratch storage and other components already left G0 and are travelling to Zellik. So far everything is developing according to plan. The clusters should arrive at their new data-center in couple hours.
In the meantime, the re-configuration of the storage system for VSC_HOME and VSC_DATA has been completed successfully. VUB users can now login to the VSC clusters of UGent and UAntwerp and from tomorrow to KU Leuven.
20:00 Transport to Zellik was uneventful. All carriages arrived at their destination in good condition.
Work started rather early today with the disconnection and proper removal of all power cords and network cables. The years of accumulated changes resulted in a bit of spaghetti to untangle in some racks.#
Network cables being sorted on the ground. The clusters have three different networks: maintenance, main Ethernet and fast InfiniBand; each one with its own type of cables. Here there are the cables of a single rack, Hydra and Anansi occupy six racks in G0.#
Everything fits in a single truck.#
All nodes are very well protected fully stuffed in foam. They’ll have a cozy trip.#
Transport carriages in building G waiting to be loaded onto the truck.#
An HPC needs many many power cords. Here lays a fraction of all power cords removed today.#
Updated on 08/09/2026
Today we got all systems placed in their new racks. In total around 80 boxes including all compute nodes of Hydra and Anansi, login nodes, scratch storage and service systems. We have also started with the cabling, which will take multiple days.
The scratch storage seems to have withstand transportation without a scratch. Upon cabling and powering it up, all drives show good health with no sign of errors. However, we still lack the network connectivity to properly check the filesystem, so we will not set this milestone as completed yet. But it is almost done.
The structure of the new data center in RPZ is peculiar. Cooling infrastructure is located in a separate room right below the computer racks and the hot air generated by the cluster is just taken by the cooling through the floor. Which is large metal grid. Hence, working with small components such as screws or cable straps can be quite nerve wracking. Items that fall through the grid have a 4-5 meter free fall.#
First step was to place all nodes in their new racks. These are the zen5_mpi with the InfiniBand NDR switch.#
We already started with the cabling once all systems where in the racks. All nodes use at least 2 power cords connected to different power delivery units (PDU) on the rack for redundancy. Left one is red, right one is blue.#
All hard drives on the scratch storage show green lights, meaning the they are healthy with no hardware errors. Which makes us very happy.#
Updated on 09/09/2026
We confirmed the good health of our scratch storage! No data is lost and we can avoid spending time restoring any backups. We just had a bit of a hiccup to get back the full redundancy across all hard drives, but by performing an obscure dancing ritual with the storage cables, we got it working.
Regarding the compute nodes, we made good progress with their cabling. All systems have power and connections to the maintenance and main networks. This allowed us to already boot-up some systems and carry out initial tests today. So far, we have not detected any hardware damage to any of the compute nodes.
Tomorrow we will complete the cabling of the fast Infiniband network, which is the last step remaining to complete the hardware installation of the clusters in the data center at RPZ.
Infiniband network cables labelled and ready to be installed.#
Kris verifying for the thousandth time the correctness of all network connections on a beautiful spreadsheet we made with many columns, and rows, and colors and alphanumeric codes. Pure art.#
Alex and Hugues installing 10 meter long optic fibers to connect the scratch storage to the racks on the other side of the corridor. Those fibers love to get tangled.#
We started powering compute nodes. The orange light means there is power but the system has not been booted up yet. These are the hopper_gpu nodes of Hydra (top) with the ada_gpu of Anansi (middle).#