We ported nine parallel programs to both TreadMarks and PVM:
Water and Barnes-Hut from the SPLASH benchmark
suite [20]; 3-D FFT, IS, and EP
from the NAS benchmarks [2]; ILINK, a
widely used genetic linkage analysis program [8];
and SOR, TSP, and QSORT.
We ran three of the nine programs using two different input sets:
Water with 288 and 1728 molecules, IS with a bucket size of
and
,
and SOR with the internal elements of the matrix initialized to either
zero or nonzero values.
The execution times for the sequential programs, without any calls
to PVM or TreadMarks, are shown in Table 1.
This table also shows the problem sizes used for each application.
Figures 1 to 12 show the speedup
curves for each of the applications.
The speedup is computed relative to the sequential program execution
times given in Table 1.
The amount of data and the number of messages sent during the
8-processor execution are shown in Table 2.
In the PVM versions, we counted the number of user-level messages and
the amount of user data sent in each run.
In TreadMarks, we counted the total number of UDP messages, and the
total amount of data communicated.