Page 17 - Oil and Gas Simulations
P. 17

APPLICATION NOTES

FIGURE 2. Left: A shared memory computer. Right: A distributed memory computer.

Figure 2).                                     FIGURE 3. Hybrid parallel computer combines both shared memory and distributed
	 Maximizing the speed of numerical            memory methods into a hybrid method.
simulations on multicore computers
relies on the fact that all cores are in use.  Figure 3).                                    One Floating Network
Shared memory computing uses threads           	 The combination of shared and               License (FNL) for COMSOL
to split up the work that needs to be          distributed memory mechanisms in              Multiphysics software
computed into several smaller work units       a hybrid method provides a versatile          allows for unlimited use
that can run in parallel within a node,        means to adapt to all kinds of computing      of CPU cores, cluster
sharing the memory. By default, COMSOL         platforms. Choosing the right way to          nodes, and simultaneous
Multiphysics utilizes all available cores in   combine the two ways of parallelization       simulation of designs.
a system, automatically ensuring that the      will speed up computations, increase
user gets the most out of it.                  scalability, and allow efficient utilization
	 In distributed memory computing,             of the hardware.
memory is distributed among several
parallel processes. These processes must
communicate with one another by
sending “messages”. As a consequence,
communication and synchronization
consumes additional time. This can be
minimized by exploiting data locality
and using improved algorithms. The
big advantage of distributed memory
computing is that it typically scales well
and additional resources (nodes, and
hence cores and memory) can be easily
added.
	 The fact that modern computers
combine both setups makes it inevitable
that both shared memory mechanisms
for intranode computations and
distributed memory mechanisms for
internode computations be exploited at
the same time.
	 The goal is to maximize scalability,
minimize the costly message passing
overhead, and utilize the power of
shared memory in a unified approach
called hybrid parallel computing (see

                                                                                             COMSOL NEWS                       17
   12   13   14   15   16   17   18   19   20   21   22