PSTSWM SPP-1200 Point-to-Point Communication Performance

Performance Studies using

PSTSWM

Convex SPP-1200 SWAP Performance

(ordered swap of 2MB message using MPI)

Date/Person: September 7, 1996 / P. Worley
Platform: Convex SPP-1200 at National Center for Supercomputer Applications (lena.ncsa.uiuc.edu):
   64 120-MHz HP PA-RISC 7200 processors (8 Hypernodes)
Environment: SPP-UX_ail 3.2 L33 01/22/96
SPP-UX_mk 3.2.144 L33 OOW elvis:/tac1/3.2.144 [CNX_MPP1]
Communication Library: MPI
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. 29.00 41.06 10.0%
10 iter. 43.73 37.49 35.9%

ordered swap using nonblocking send
Data Statistics
unidirectional bandwidth estimated latency model error
(peak MByte/sec) (usec/msg) (max. rel. error)
1 iter. 28.91 39.67 3.2%
10 iter. 43.61 42.87 35.9%
1 iter. w/overlap 60.88 41.14 12.5%
10 iter. w/overlap 80.91 42.45 37.0%

ordered swap using nonblocking receive
Data Statistics
unidirectional bandwidth estimated latency model error
(peak MByte/sec) (usec/msg) (max. rel. error)
1 iter. 28.95 38.73 3.8%
10 iter. 43.85 40.90 36.2%
1 iter. w/overlap 61.05 42.52 14.3%
10 iter. w/overlap 79.60 42.17 35.0%

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. 29.05 26.97 13.5%
10 iter. 43.60 41.42 35.9%
1 iter. w/overlap 60.80 48.33 8.8%
10 iter. w/overlap 79.41 47.19 35.3%

ordered swap using ready send
Data Statistics
unidirectional bandwidth estimated latency model error
(peak MByte/sec) (usec/msg) (max. rel. error)
1 iter. 28.92 58.75 3.3%
10 iter. 43.34 55.19 35.6%
1 iter. w/overlap 60.92 46.25 9.1%
10 iter. w/overlap 86.73 42.70 36.0%

ordered swap using nonblocking ready send
Data Statistics
unidirectional bandwidth estimated latency model error
(peak MByte/sec) (usec/msg) (max. rel. error)
1 iter. 28.90 59.35 7.1%
10 iter. 43.33 60.20 35.8%
1 iter. w/overlap 60.77 53.52 8.8%
10 iter. w/overlap 85.99 48.36 35.2%

synchronous
Data Statistics
unidirectional bandwidth estimated latency model error
(peak MByte/sec) (usec/msg) (max. rel. error)
1 iter. 28.99 44.75 6.7%
10 iter. 43.72 53.70 35.8%

DISCUSSION

Patrick H. Worley / ( worleyph@ornl.gov)
Last Modified Monday, 15-Jul-2002 10:22:43 EDT.
86778 accesses since 1/2/96.