DOMA / QoS Meeting

Paul Millar (DESY)

Discussion on timings and mandate:

  • Next meeting to be doodled,
  • 2 week rhythm too much for some participants

Mario is our ATLAS contact
Martin is our Rucio contact

We should state a connection to workload management systems.  For example, exploiting low reliability storage opportunistically and not tracking its contents in a catalogue.

QoS example :  heavy ion jobs need lots of memory and need high memory queues.
Data has to go to those few data centres with high memory queues.  What's the cost of moving this data?  Can alternative QoSs help?

Connection to hi-lumi and meeting those requirements

There was a feeling we need competition in various dimensions (capacity, iops, reliability) which must therefore be exposed.  Then the experiments can choose the right (i.e. cheapest) storage for a particular job.

# activities and scope

Out of scope:

  • benchmarking analysis on SSDs
  • when does it make sense for an experiment to use SSDs -> access WG (?)

    but QoS WG needs the results of such a study

In scope:

  • declaring (kind of QoS of given storage capacity),
  • discovering (what is available QoS options),
  • quantifying, modifying, ...
There are minutes attached to this event. Show them.