Performance Studies using PSTSWM

Intel Paragon Protocol Performance

(transpose LT experiment I-A1 / O(P) sendrecv transpose algorithm)

1x16 Processors / Problem T10L16

Runtime Statistics min (mean-min)/min (median-min)/min (max-min)/min   4.5168e-01    0.14    0.11    0.33

Three Fastest Protocols 1st 2nd 3rd   c2    c3    a1         Number of Proctocols With Runtimes Within X% of Min 1% 5% 25%   1    2    22

1x8 Processors / Problem T21L8

Runtime Statistics min (mean-min)/min (median-min)/min (max-min)/min   1.0770e+00    0.03    0.02    0.20

Three Fastest Protocols 1st 2nd 3rd   c2    d2    c3         Number of Proctocols With Runtimes Within X% of Min 1% 5% 25%   6    23    25

1x16 Processors / Problem T21L16

Runtime Statistics min (mean-min)/min (median-min)/min (max-min)/min   1.1935e+00    0.05    0.04    0.22

Three Fastest Protocols 1st 2nd 3rd   c2    a3    a2         Number of Proctocols With Runtimes Within X% of Min 1% 5% 25%   5    15    25

1x32 Processors / Problem T21L32

Runtime Statistics min (mean-min)/min (median-min)/min (max-min)/min   1.5200e+00    0.08    0.07    0.33

Three Fastest Protocols 1st 2nd 3rd   a2    c2    b0         Number of Proctocols With Runtimes Within X% of Min 1% 5% 25%   5    15    29

1x8 Processors / Problem T42L16

Runtime Statistics min (mean-min)/min (median-min)/min (max-min)/min   9.7947e+00    0.02    0.02    0.19

Three Fastest Protocols 1st 2nd 3rd   d6    b6    b4         Number of Proctocols With Runtimes Within X% of Min 1% 5% 25%   10    24    25

DISCUSSION

Partitions of the Paragon processor grid were used that match the processor subsets that the parallel algorithms would run on in a two dimensional data decomposition: 1x8, 1x16, and 1x32.

---

Patrick H. Worley / ( worleyph@ornl.gov)
Last Modified Monday, 15-Jul-2002 10:30:04 EDT.