Table 21 shows the efficiency of the MecaGRID using 64 processors.
In Table 21, T1 is the computational time, T2 the local communication time,
and T3 the global communication time. Eff, R2, and R3 are the computational,
local communication, and global communication times relative
to the globus run using 32 nina and 32 pf processors.
As seen in Table 21, applications using 64 CPUs are in the range
4-5 times slower than the same run on the INRIA cluster.
The reader is referred to APPENDIX G for additional discussion.
Table 21:
1.03M mesh: Efficiency using 64 CPUs with Sav = 5