Information Technology Reference
In-Depth Information
Fig. 6.
Vectorial form of the ODE system in a shared memory
number of terms per core, which resulted in a fair balance. The allocation of
terms uses a fine-grained round-robin scheduling.
Since the matrix
ˁ
is a strongly coupled system, it should be stored in a shared
memory to allow that every core can read the vector
ˁ
(
i,j
) simultaneously
13
.
6 Experimental Setup
Three models of QDsim were implemented to analyze its performance under
different environments and applying two partition schemes (LP and TP):
-
A shared memory model using OpenMP
14
-
A distributed memory model using OpenMPI
15
-
A hybrid model (distributed/shared memory) using OpenMPI & OpenMP
Basically, these versions tried to answer the following questions:
1. Has a shared memory architecture the same performance as a distributed
memory architecture?
2. Does hyper-threading technology contributes in the computing performance?
13
The simultaneity depends on the size of the level 2 cache and the data size to be
mapped into it.