This session primarily covers aspects of production (skim, PR, SP) plus (if I finish a presentation about it) a talk about how we specify the link dependencies WRT ROOT libraries for our executables. I'll concentrate on the production parts here. Skim production --------------- The CMWG2 requirements document has two relevant requirements: 5.5.1 Both Mini or New Micro format skims should be supported, with the computing and physics management responsible for determining which skims are in which formats. 5.5.2 Either pointer or deep copy collections should be supported, again dependent on the size of the output and the needs of the AWG. Skim production: o Should we separate out PhysProdSequence from Elf/Moose? i.e. Should Elf/Moose produce only Mini and AllEvents Micro? o What is the tag in that case? o Assuming we still produce it immediately (in OPR/REP), what is the policy? (Latest executable used in a every 3 month sweep?) o Or wait for up to 3 month latency to produce physics files? o What is the current speed of skim production executable? o How to deal with customized skim definition? o Beyond choosing lists for which we write composites, what other user info is written into customized skims? o How is code provided for that? o Is there a cpu budget for this? o (Namespace in transient event for non-list information to write?) o What is supported? o Writing pointer skim o Writing deep copy micro o Writing deep copy micro and composites o Writing deep copy micro, composites and other "numbers" o Writing a pointer to AllEvents micro event and "user" data o How do we deal with (small) file size? Should we merge? Are there any problems with doing this when good runs list is changing? If we merge, what are the collection naming conventions? o Who owns the skim events? (i.e. is it possible for AWG's to clean up there own mess or does it require an administrator? By "clean up" here I mean in particular make something non disk-resident, I assume that this will not be possible if it does not reside on tape.) o How do we do the tcl configuration (by skim) plus the "opt-in" skim specification for the new skim executable? o How much memory is required to write 100 (customized) skims from a single executable?