Page History

...

• run a batch job in s3df
• get real-time plots
  • for the latest run by default
  • if they go reanalyze old runs, that would be a user-selectable option
• could use X-forwarding from s3df to start, but ideally would use open-ports we requested (latter may be stalled)
• timescale: 2 months would be nice, 3 or 4 months is probably OK

how:

• psmon (zmq+pyqtgraph).  eventually use “psplot” command-line to display the plots.
• murali’s elog database to learn which SRV node/port is serving the plots (most challenging)
• to start require only 1 SRV node
• if we eventually need more than 1 SRV node we would have to write another layer on top of SRV with 1 core collecting and serving the plots
• to make it easy for scientists:
  • need some sort of script to poll the elog database for the most recent run (most challenging)
  • if a run number goes backwards ignore it by defaults

...

• need to get "mpirun -n 5 python andor.py" working (psmon issue).  hack of "if rank==4: publish.init()" seems to work.  then can eliminate "publish.local=True" and use "psplot ANDOR" to get plots
• two shmem issues (why we use live-mode, not shmem):
  • easy to drop events to shmem (don't want this for normalization)
  • hard for ric's event builder to send groups of events reliably to one core of the shmem
• one offline issue:
  • can miss events due to deadtime but Matt provides a counter so experimenters can watch for when this happens (see above code example)