Top Issues in ROOT and CMSSW I/O
These refer to pure I/O, and ignore the issues in the serialization/deserialization of CMS data structures.
- Raw TTreeCache: For the first cluster of TTree, read in all branches by default.
- Align learning periods in CMSSW to cluster boundaries.
- Don't reset learning in multicore mode for every event skip.
- Ability to add all branches to TTreeCache from config file. Note: with the exception of the high-latency case, there's almost no standard use case where reading the whole event doesn't make sense.
- Ability to manually set the flush interval for RECO/AOD tiers. Will help mitigate the fast merge issues for these files.
- Repeat all improvements for FWLite.
- Double-buffering in TTreeCache (under development by CERN).
- Use TTreeCache-based I/O in fast merge.
- Fix statistics for native ROOT plugins.
- Fix fast merge of files with different flush intervals.
- Upstream raw TTreeCache into ROOT: Align training period length to cluster boundaries. For the first cluster of TTreeCache usage, read in all branches.
- TAdaptiveTreeCache: keep usage statistics from TBasket, and update the list of branches once every cluster transition.