What is the best way to access UPC events?

[bdrum]

In the slides of the one presentation I've seen:

We have two types of UPC events Triggered by MUON Triggered in the Central Barrel In both cases the events are almost empty The average size of AOD UPC events in the central barrel for 2015 PbPb is around 0.5 kB We have some 30 M triggered events, which means some 15 GB of data. The purity of this sample is small, meaning that a simple pre-selection may reduce this number by a large factor (between 2 and 3).

In the past we have used a LEGO train to access the events and produced trees to perform local analyses. The code in the LEGO train is quite stable Latchezar commented that this approach may not be optimal for such small data set distributed over all the PbPb data set and recommended that we had a look at nanoAODs.

The code to select only those branches we use and create with them ‘a nanoAOD’ is ready. As it is so general, we do not expect the code to be changed frequently. Latchezar mentioned that there exist AOD trains in the central framework.

Could it be a good option to use them to produce UPC AODs?

In Summary, what is the best way to run over the UPC triggers in the PbPb sample so that we do not have to run over all the PbPb events to find our few gigas of data?