PSTSWM SP2-66 Point-to-Point Communication Performance

Performance Studies using

PSTSWM


IBM SP2-66 SWAP Performance

(ordered swap of 2MB message using MPI)

Date/Person: January 13, 1995 / B. Toonen
Platform: IBM SP2 at NASA Ames Research Center (babbage.nas.nasa.gov):
     160 RS6000/590 nodes ("wide", 66.7 MHz POWER2 RISC chip set)
Environment: AIX
Communication Library: MPI-F version 1.3.8
SWAP size: 262144 REAL*8 floating point values each direction
Message size: Largest - 262144 REAL*8 floating point values
Smallest - 256 REAL*8 floating point values
Processors: 0 and 1
Latency Definition:(T512-T256)/256
Results:

ordered simple swap
Data Statistics
unidirectional bandwidth estimated latency model error
(peak MByte/sec) (usec/msg) (max. rel. error)
1 iter. 70.78 73.09 23.3%
10 iter. 70.70 69.82 22.7%

ordered swap using nonblocking send
Data Statistics
unidirectional bandwidth estimated latency model error
(peak MByte/sec) (usec/msg) (max. rel. error)
1 iter. 70.81 73.80 23.2%
10 iter. 70.65 72.27 22.0%
1 iter. w/overlap 70.24 76.81 24.0%
10 iter. w/overlap 70.61 73.11 23.6%

ordered swap using nonblocking receive
Data Statistics
unidirectional bandwidth estimated latency model error
(peak MByte/sec) (usec/msg) (max. rel. error)
1 iter. 70.78 75.66 23.5%
10 iter. 70.40 76.57 21.3%
1 iter. w/overlap 70.72 73.88 26.3%
10 iter. w/overlap 70.46 74.58 23.3%

ordered swap using nonblocking send and receive
Data Statistics
unidirectional bandwidth estimated latency model error
(peak MByte/sec) (usec/msg) (max. rel. error)
1 iter. 70.42 77.08 22.6%
10 iter. 70.74 76.64 21.7%
1 iter. w/overlap 70.76 83.74 24.4%
10 iter. w/overlap 70.37 79.11 22.6%

ordered swap using ready send
Data Statistics
unidirectional bandwidth estimated latency model error
(peak MByte/sec) (usec/msg) (max. rel. error)
1 iter. 70.61 83.24 7.5%
10 iter. 70.76 81.95 6.0%
1 iter. w/overlap 70.82 79.96 9.5%
10 iter. w/overlap 70.40 75.06 7.8%

ordered swap using nonblocking ready send
Data Statistics
unidirectional bandwidth estimated latency model error
(peak MByte/sec) (usec/msg) (max. rel. error)
1 iter. 70.81 80.34 9.6%
10 iter. 70.81 80.01 6.5%
1 iter. w/overlap 70.94 82.82 9.8%
10 iter. w/overlap 70.80 79.74 8.5%

synchronous
Data Statistics
unidirectional bandwidth estimated latency model error
(peak MByte/sec) (usec/msg) (max. rel. error)
1 iter. 70.67 76.10 23.7%
10 iter. 70.57 77.67 21.9%


Patrick H. Worley / ( worleyph@ornl.gov)
Last Modified Monday, 15-Jul-2002 10:13:57 EDT.
86808 accesses since 1/2/96.