I've turned off the auto mode of the pileup analyzer because:
it was taking too long (need to read the first file of each component through xrootd)
it would have been difficult to reach perfect sync with the other groups.
Instead, we are now using a root file for MC pileup distributions provided by Artur. The file contains a distribution for each MC dataset. To be able to use this feature, you need the new version of the PileUpAnalyzer, so you need to fetch again https://github.com/cbernet/cmg-cmssw/tree/94X_HTT
In mu-tau I removed the production option and added a test option instead. when activated, you will read interactively the 1st file of VBF H125. I suggest to do the same for the other channels.
I investigated the long startup times of heppy
a lot of time is spent finding out how many events there are in a dataset. heppy does this using the python fwlite interface, which probably opens all files and sums up the entries... so when debugging and working interactively, it is important to reduce the number of files in the input component. The test option I have implemented for mu-tau does this.
event processing speed is increasing over time. This seems to be due to xrootd read speed. When working interactively, just copy one input file from VBF H125 locally with xrootd, and change your configuration file to read it.
Finally, I started to clean up our components to be able to read the VBF H125 sample easily. @GaelTouquet, you will need to merge this branch before your PR this morning.
Hi @steggema @GaelTouquet @lucastorterotot,
I've turned off the auto mode of the pileup analyzer because:
Instead, we are now using a root file for MC pileup distributions provided by Artur. The file contains a distribution for each MC dataset. To be able to use this feature, you need the new version of the PileUpAnalyzer, so you need to fetch again https://github.com/cbernet/cmg-cmssw/tree/94X_HTT
In mu-tau I removed the production option and added a test option instead. when activated, you will read interactively the 1st file of VBF H125. I suggest to do the same for the other channels.
I investigated the long startup times of heppy
a lot of time is spent finding out how many events there are in a dataset. heppy does this using the python fwlite interface, which probably opens all files and sums up the entries... so when debugging and working interactively, it is important to reduce the number of files in the input component. The test option I have implemented for mu-tau does this.
event processing speed is increasing over time. This seems to be due to xrootd read speed. When working interactively, just copy one input file from VBF H125 locally with xrootd, and change your configuration file to read it.
Finally, I started to clean up our components to be able to read the VBF H125 sample easily. @GaelTouquet, you will need to merge this branch before your PR this morning.
Cheers!
Colin