There is no hard limit on how many jobs can be handled by a single Submit node. The load balancing between the nodes it done based on CPU load which is being regularly checked and recorded in the HC DB. However a lot of tuning (work distribution) is one manually, therefore continues level of intervention is required for operations. The main problem is memory. If the test is running for a long while (not just one day), the Submit node can easily run out of memory. Work around is required in such cases.

Too many layers for a given test test to reach a given resource

Considering components shown on the schema, there is no components which represent shared instances between several experiments, there is either a dedicated instance per experiment, or multiple instances dedicated to a particular experiment

Server Node and Submission node are coordinating through the HC DB

Job tracking in GANGA is implemented in sqllite. There is also job tracking info recorded in the HC DB and synchronized with the GANGA one. HC for scheduling purposes does not use the GANGA job repository.

Interaction with GANGA is done through files, not through APIs.

Code is commited to the head, no branches. There are about 5 people who can commit.

Need to better understand and agree on the HC organizational issues. Will contact Johannes and other ATLAS HC colleagues.

There are two topics for the next meeting : 'HC as a service' and closer look at test configuration/generation/submission

In parallel will start filling 'requirements and use-cases' part (present and future)

-- JuliaAndreeva - 17 Apr 2014