End of Brutus
All good things come to an end.
After over 9 years of operation, Brutus is approaching the end of its life and will be shut down forever on 16 January 2017.
Why shut it down?
There are in fact several reasons to shut Brutus down:
- First of all, most of the cluster's hardware is very old: its oldest compute nodes were purchased between 2007 and 2009, and the newest ones between 2011 and 2012. The vast majority of this hardware is no longer under warranty. Keeping such old hardware up and running requires a lot of maintenance work and is expensive from a power and cooling point of view.
- Its infrastructure is old too. The water-cooled racks where the compute nodes are installed are way past their prime. Getting spare parts is becoming so difficult that it is simply not possible to keep this infrastructure in operation much longer. The people in charge of the infrastructure have therefore decided to decommission these water-cooled racks and the underlying infrastructure — heat exchangers, pumps, pipes, etc. — at the beginning of 2017.
- According to the "RZ-Strategie" of ETH, the main computer room in the RZ building will be renovated and reserved for mission-critical systems. HPC clusters like Brutus do not fall in this category. Its removal will make room for more important servers.
Last but not least, our limited manpower does not allow us to build new clusters and offer new services while maintaining old clusters at the same time. Shutting down Brutus will allow us to expand and improve the Euler cluster, build the new Leonhard cluster, and tackle a wider range of scientific applications than ever.
When and how will this happen?
As mentioned above, the cluster will remain in operation until the end of the year.
- We will stop creating new user accounts on 31.10.2016.
- The cluster will be shut down permanently at noon, Monday 16 January 2017.
- As required by data privacy law, all data stored in the cluster will be deleted before the corresponding storage systems are shut down and decommissioned. This includes all data stored in "home", "work" and "scratch" file systems. Please keep in mind that "work" and "scratch" directories are NOT backed up. The last tape backups of "home" directories will be kept as usual (2 monthly backups) and will be deleted permanently in March 2017 when the last monthly backup expires. We make not guarantee whatsoever regarding the possibility to retrieve data from these backups after the "home" file servers have been shut down.
The safekeeping of your data is your responsibility. All users are kindly requested to copy the data that they want to keep to another system before the cluster is shut down. If you do not know where to store these data, please contact us; we will try to find a solution.
Where can I compute once Brutus is shut down?
The end of Brutus does not mean the end of HPC at ETH — quite the contrary:
- The Euler cluster, which has been in operation since the beginning of 2014, already offers 5 times the computing capacity of Brutus. Its upcoming expansion — Euler III — later this year will offer much faster processors and more computing capacity than the whole of Brutus. Although it uses the same "shareholder model" as Brutus, only those groups who invested in Euler own a dedicated share of this cluster. Brutus shareholders who have not (yet) invested in Euler are considered as guests on this cluster.
- The future Leonhard cluster, which has been recently approved by the Executive Board of ETH, will be installed at the beginning of 2017 in a whole new computer room located in the LEE building. Leonhard will be aimed at "big data" applications and will offer very large and fast storage systems, combined with high-throughput compute nodes as well as GPU nodes of the newest generation. Leonhard will also offer enhanced security for confidential applications like personalised medicine or finance.
The majority of Brutus shareholders have already switched their computations to Euler. Those who have not can do so at any time or, depending on their applications, invest in the new Leonhard cluster at the beginning of 2017. We remain at your disposal to analyse your requirements and find the best solution with you.
