Memory overhead caused by case-insensitive lookup in the JobOptionsSvc in ATLAS
Hi all,
In the master branch of Athena, we're using a patched version of Gaudi v34 (v34r0.005). Looking at the memory usage in a standard reconstruction job, we've noticed that the case-insensitive lookup infrastructure of the new JobOptionsSvc is using multiple hundreds of MBs of memory:
In many cases, we have tens of thousands of property clients. In addition, we're using fully qualified names, e.g. Alg.Tool1.Tool2.Tool3.Prop
, which can be really long. As a result, it seems as if the internal maps end up using a significant amount of memory.
Let me cc @ssnyder and @fwinkl since I'm sure they have valuable input/ideas on this topic at large. However, I think it would suffice to say that it would be great if we can optimize things on this front.
Please let me know, many thanks.
Best, Serhan