Here are some topics which were mentioned during the first meeting and we may want to discuss them further. Please feel free to edit the list:
- What do we want to share, document and keep up to date?
- taxonomy/ontology of jobs?
- questions being explored?
- tools being used?
- available data? data schemas? level of trust?
- just links to various sources?
- How do we want to share it?
- GitBook?
- One place
- Ability to keep it up to date
- What exactly are we trying to optimize for?
- Try to agree on a common metric of efficiency based on throughput and cost as mentioned by Markus Schulz in his presentation on The 'Understanding Performance' Team https://indico.cern.ch/event/539739/
Taken decisions:
- We want to start sharing (and update at least once a month):
- taxonomy/ontology of jobs as it will be continually revealed
- who is currently exploring which questions with which tools
- We want to use TWiki for that.
- We will set up a sub-e-group of the AWG e-group for this meeting.
Discussion about common metric for efficiency:
- At least ATLAS should have a standard set of jobs referred to as "standard candles", which could be used for benchmarking, experimenting with various HW configurations and exploring effects of running different kinds of candles on one machine at the same time.
- Marcus mentioned that there is a need for a controlled reference system and suggested that T0 reconstruction jobs can be used on OpenLab machines to start work on establishing this reference system.
Next meeting to be chaired by Andrea Sciaba.
There are minutes attached to this event.
Show them.