CERN Accelerating science

Optimal Scheduling Group Config


(John White) #1

I have a new EOS instance and am trying to figure out the optimal scheduling group configuration. I have 2 FSTs with 20 individual file system disks each. I slice off 1 FS each to a ‘spare’ group and (currently) have the other 38 in a single scheduling group. The reasoning here is to let the scheduler just go buck wild and balance across everything.

Could I get some discussion about the possible disadvantages of this?


(Franck Eyraud) #2

Hi John,

The standard suggested group layout, that we use in our production environment, is to have in each group FS that are hosted by different FSTs. So in your case you would probably have 20 groups with 2 FS in each, or 19 groups and keep 2 spare FS.

This is useful for the standard file layout with 2 replicas : a file will be placed in one group, each replica on two different FSTs, so if one FST is down, the file is still accessible.

But if you plan to store your files with plain layout of only one replica, then this observation probably doesn’t apply to you, you’ll have a less available file system because of a lack of redundancy, but a larger available space, and I suppose using one group doesn’t change much. But think of it if you plan to extend the storage in a later stage.

But maybe someone from eos development team might have more precise suggestions.

Hope this helps,

Franck