Compute jobs on CIP-Workstations

What is this about?

The usage by students and employees of the compute power provided by the CIP-Pool Workstations has risen sharply. To ensure an orderly usage and to ensure that normal opperations due not suffer some rules of conduct are to be observed.

Why does this need regulation?

Our sytem is designed for a medium load with occasional utilisation peaks. This is currently the best configuration because of cost and effort reasons. If for example you have a study assignment and start a script on dozens of machines and that script reads and/or writes lots of data to/from your homedir you will almost invariably overload the network connection of the room and the fileserver for your homedir (and the homes of lots of other users). Furthermore this type of usage is counterproductive because it slows down your own jobs compared to a comprehensibly planned resource usage scenario.

Another rule is that the user physically sitting in front of the machine always gets priority in using the machine.

The SSH and RDP computers, computers reachable from outside of the CIP-Pool in other words, are to be kept free of any kind of compute usage.

What do I need to do then?

In case of doubt always ask the RBG !!!

In any case:

Priority for interactive users

If someone is using the computer interactively you may not start a compute job. Running jobs need to be suspended. The interactive usage has priority in every case. Using SLURM satisfy this constraint.

Prevention of high network load

You may not run programms which create excessive load on the local network or beyond. Data traffic inside the institute: Do not use to much bandwidth. In case of doubt ask the RBG. Data traffic outside the institute: Here we must differentiate between traffic to the MWN and to the rest of the world. The pools are connected to the MWN by 1Gbit links. You should never use more than 10% of the bandwidth. If the traffic goes beyond the MWN you need to worry about the transfered amount of data as well as an even lower amount of bandwidth. You have to get the RBGs approval beforehand if more than 4 Gigabytes per day need to be transfered because this ammount of data will trigger the traffic monitors of the LRZ. Using the pools at night is prefered.

Dont use the our ssh/remote pool for compute jobs

Dont start compute jobs or tests of jobs on this pool if the job duration is longer than a couple of minutes. Use the remote pool as a gateway to normal CIP computers. Submitting of SLURM jobs is allowed.

No longterm data intensive usage of your homedir

Our fileservers are not configured to support longterm data intensive usage. For this the ratio of clients to servers as well as the ratio of homedirs to servers are to high. The result of such an usage is that all users of a server overloaded like this can only work very slowly.

It is therefore easy to use the local disks of a CIP-Pool Workstation. This usage is recomended anyway because the local harddisks are significantly faster than the homedirs which need to work through the network.

In case of jobs running for multiple days

The RBG is to be informed of such usage beforehand in any case. This is best done via email to rbg@ifi.lmu.de. In this Mail you need to explain what you want to compute and the span of time you think it will take. Only jobs which relate to your courses are permitted.

What happens in case of violations?


Links

External Infos