60 likes | 152 Views
DAQ-HLT discussion. Dataflow is simpler in the sense that there is no more decisions to pass. Single stream although there is the buffering (cont. r.o.) to complicate things.
E N D
DAQ-HLT discussion • Dataflow is simpler in the sense that there is no more decisions to pass. Single stream although there is the buffering (cont. r.o.) to complicate things. • HLT does no actual trigger, but compression. DAQ provides at different stages data to HLT « boxes » (processes/threads) and gets less out. • The factor 5 compression at FLP stage can be achieved with cluster finding either SW or HW or both • The rest of the HLT work happens on EPNs
DAQ-HLT discussion • FLPs and EPNs are concepts and might or might not be running on different machines. components must be able to run on a single machine as well as multiple machines pluggable system to decide what protocol to use • It is also useful for the ability to test components together or in isolation
DAQ-HLT discussion • Threads vs processes : what to use when ? Could we use only threads ? We need common frameworks for that. • Do not develop what already exists (inside or outside the community) • Messaging system ? • Load balancing ?
DAQ-HLT discussion • 12 hours of data locally recorded: feasability?Could we store some data on the EPNs disks? • Fault toleranceBe more flexible than now to losing events or machines. Mask out a broken FLP during run. Use a different EPN on the fly. dynamic configuration
How we see our role • Strategy group • Define the strategy, the architecture, the guidelines, the responsibilities of the future ALICE software (analysis, calib, …) • Question, dispute and challenge the LOI (and other documents if relevant) • Write a document and present it
Where to go from here • Raise questions and concernsDon’t answer or solve as we don’t have time • Based on the conceptual data flow, build a more detailed view (boxes and pipes) -> first definition of packages and responsibilities • List all this in a document • Present • Create working groups to work on points listed in the document