Three presentations were given at the November meeting.
Andreas Hirstius updated us on the
server situation after the physical move in the Computer Centre.
He highlighted the brand new HP/Intel blade configuration which
gives us ample capacity for benchmarking, workshops, compiler
testing and the like.
Hċvard Bjerke gave an update on the
track finder benchmark that he is developing together with
Intel, Brühl and KIP/U. of Heidelberg. The approach exploits
parallelism to the highest degree possible by using a derivative
of the Cellular Atomaton algorithm.
He concluded by saying that the track finder has now been
integrated with the AliHLT framework.
Jeff Arnold, who is a
visitor from the Intel Compiler labs in the US, discussed his
finding when
using process forking inside the LHCb reconstruction framework.
He showed, based on his tests, that such an approach can reduce
memory consumption by ~50% with the advantage that no internal
application changes are needed.
Sverre Jarp
CERN
|