Next: User Guidelines
Up: Introduction
Previous: Introduction
The model generator program lets the user adjust the following parameters
before writing out the model:
- (a) Number of processors or caches to simulate - This is the number
of SMMP processors that are required to be simulated. Each SMMP processor
is represented by a set of queuing objects that perform the task of
symmetric multiprocessor. In our model, a processor along with its cache
memory is modeled with six queuing objects (Source, Join, Queue, Fork,
Server and another Join). Figure 1.1 illustrates the complete
set of queuing objects that are required for modeling a 8 processor
symmetric multiprocessor.
- (b) A Sequential or a Parallel Simulation - The WARPED
simulation kernel can be deployed as a parallel simulation kernel as well
as a sequential kernel.
- (c) Number of physical processors (or Logical Processors) - This is
the number of processors that the user wants to use on the host machine
for a parallel simulation. The generator program requires that the number
of processors or caches (i.e., option (a)) be an integral multiple
of the number of physical processors. This is required to create correct
partitions.
- (d) Speed of cache - This is a model parameter that needs to be set
by the user. We assume that the speed of the cache is an integer quantity
which is smaller than the speed of the main memory. A typical value for
the speed of the cache would be 10, whereas the speed of the main memory
can be 100, implying that the cache is 10 times faster than the main
memory. The main objective here is to model the latencies of the cache and
the main memory.
- (e) Cache Hit ratio - again this is another model parameter that
determines what percentage of the requests are sent to the cache memory
instead of the main memory. Obviously, the lower this number is, the more
requests that are sent out to the main memory ( i.e., more cache
misses).
- (f) Speed of main memory - Basically the latency of the main memory
represented as a multiple of the cache speed. Again we assume here that
this is an integer quantity.
- (g) Number of requests per processor - This is the number of memory
requests each processor with generate during the lifetime of the
simulation.
- (h) Name of file to generate - The generator will write out the
model to the file name specified by the user. By default, the generator
program will write out the model to main.cc.
Figure 1.1: A 8 processor SMMP model
Next: User Guidelines
Up: Introduction
Previous: Introduction
Radharamanan Radhakrishnan
Fri Mar 12 00:05:34 EST 1999