Geoscience Reference
In-Depth Information
8 Handlers
16 Handlers
9
8
7
6
5
4
3
2
1
0
0
2
4
6
8
Number of Threads
Fig. 9.6
Performance of parallel event handling
is made. Even more complex parallel programs written in C CC with MPI achieve
speedups of around 8 in this case (Charypar et al. 2007a ).
While only a small number of event handlers are present by default in MATSim,
many applications are under development and planned in the MATSim community
(MATSim 2009 ) which require additional event handlers. The good news is that if
more handlers are added to parallel event handling, the speedup gets slightly better,
as can be seen in the 8-thread scenario. This is expected because adding more work
to the handler reduces the relative penalty of synchronization between threads.
9.4.4
Speedup for Parallel JDEQSim
As described earlier, a first prototype of the parallelization of JDEQSim for
two threads has been implemented. As event handling has not been adapted yet
to properly function with parallel JDEQSim, only measurements of the micro-
simulation are reported here where event handling is turned off. The experiment
consists of 1.62 million agents residing in the surroundings of Zurich city. The
network contains 163 K links. This experiment required 29 min and 40 s when
running with JDEQSim, while on the parallel JDEQSim (2 threads), the experiment
only took 18 min and 37 s. This is a speedup of 1.6, which is encouraging, but many
problems remain unresolved, as is discussed in the future work section.
Search WWH ::




Custom Search