February maintenance window open
The February maintenance window will begin at 09:00 CET on Wednesday 4 February.
-
Login nodes will be restarted.
-
All systems will receive important bug fixes and security updates.
-
The queues will be stopped for Pelle.
-
Rackham and Snowy will be shut down the last of January. So at the time of the service window they are not with us anymore.
-
We will move the last servers with Nvidia T4 GPUs from Snowy to Pelle and put them in the haswell partition.
Please use the haswell partition for only GPU jobs for now.
Read more about how to start jobs on Pelle with GPUs
In the future we plan to use the haswell partition as a surge resource. If all cores on Pelle are full – maybe it is better to run on a slower core than not being able to run at all. We do recommend running on Pelle first.
-
We will start moving projects from Crex to Gorilla, one by one and in groups.
-
We will start doing migration of projects from the old system Crex to the new system Gorilla. We will do this project per project and send mail to users when their data is being moved since this will change how the data is accessed by changing the file system paths.
Please note that this will for all users mean new paths to where the data is located in their project directory. We have discussed a lot about how to do this in a way that is reasonably logical. We will also stop defaulting on backing up everything and instead backing up only the backup directory for every project. In many cases many smaller projects will be collected into one project.
So before the move:
/proj/oldprojectname (backed up)
/proj/oldprojectname/nobackup (not backed up)
After the move:
/proj/newprojectname/ (not backed up)
/proj/newprojectname/backup (backed up)
/proj/newprojectname/oldprojectname1 (not backed up, all data from oldprojectname1)
/proj/newprojectname/oldprojectname2 (not backed up, all data from oldprojectname2)
/proj/oldprojectname (will NOT work)
/crex/proj/oldprojectname (will NOT work)
We sincerely hope this will not be too confusing and we will make our best to make the transition as painless as possible.
-
We are doing maintenance on the clusters preparing to take parts of the storage system Gorilla in production. We plan to sync datasets which reside in the /sw/data on Crex to the new storage system Gorilla. In order to do that all nodes in Pelle, Rackham and Snowy will be stopped during the day. We plan to bring them up again later during the day.
-
Other than this, no changes are planned. But as always, there is an increased probability of minor glitches or outages.
-
Please follow progress on https://status.uppmax.uu.se.