PSTSWM SPP-2000 Point-to-Point Communication Performance

Performance Studies using

PSTSWM

HP/Convex SPP-2000 SWAP Performance

(unordered swap of 8KB message using MPI)

Date/Person:	April 26, 1998 / P. Worley
Platform:	HP/Convex SPP-2000 at National Center for Supercomputer Applications (billie.ncsa.uiuc.edu):
	64 180-MHz HP PA-RISC 8000 processors (4 Hypernodes)
Environment:	SPP-UX_ail 5.2.1 L34
	SPP-UX_mk 5.2.1.139
	HP MPI 01.03.01.00 (11/21/97) B5880 - SPP-UX 5.x
Communication Library:	MPI
SWAP size:	1024 REAL*8 floating point values each direction
Message size:	Largest - 1024 REAL*8 floating point values
	Smallest - 1 REAL*8 floating point value
Processors:	0 and 1
Latency Definition:	(T₁₀₂₄-T₅₁₂)/512
Results:

unordered simple swap

Data Statistics
bidirectional bandwidth estimated latency model error

(peak MByte/sec) (usec/msg) (max. rel. error)

1 iter. 85.31 20.14 87.3%

10 iter. 53.84 18.89 38.0%

unordered swap using nonblocking send

Data Statistics
bidirectional bandwidth estimated latency model error

(peak MByte/sec) (usec/msg) (max. rel. error)

1 iter. 134.22 13.26 36.1%

10 iter. 182.84 13.23 27.4%

1 iter. w/overlap 132.15 15.76 36.0%

10 iter. w/overlap 178.86 14.34 29.4%

unordered swap using nonblocking receive

Data Statistics
bidirectional bandwidth estimated latency model error

(peak MByte/sec) (usec/msg) (max. rel. error)

1 iter. 144.98 22.88 23.5%

10 iter. 179.05 23.11 25.3%

1 iter. w/overlap 157.61 23.79 24.4%

10 iter. w/overlap 198.61 23.02 27.9%

unordered swap using nonblocking send and receive

Data Statistics
bidirectional bandwidth estimated latency model error

(peak MByte/sec) (usec/msg) (max. rel. error)

1 iter. 144.98 22.60 24.3%

10 iter. 187.89 23.38 26.8%

1 iter. w/overlap 151.70 27.52 25.5%

10 iter. w/overlap 192.09 25.51 29.9%

unordered swap using ready send

Data Statistics
bidirectional bandwidth estimated latency model error

(peak MByte/sec) (usec/msg) (max. rel. error)

1 iter. 125.06 39.72 30.3%

10 iter. 154.86 39.97 37.8%

1 iter. w/overlap 157.61 24.63 23.7%

10 iter. w/overlap 200.29 22.64 27.7%

unordered swap using nonblocking ready send

Data Statistics
bidirectional bandwidth estimated latency model error

(peak MByte/sec) (usec/msg) (max. rel. error)

1 iter. 124.15 40.38 30.6%

10 iter. 157.85 39.73 38.3%

1 iter. w/overlap 149.07 27.14 24.7%

10 iter. w/overlap 191.85 25.48 29.8%

native sendrecv

Data Statistics
bidirectional bandwidth estimated latency model error

(peak MByte/sec) (usec/msg) (max. rel. error)

1 iter. 140.10 14.36 28.5%

10 iter. 182.45 13.85 29.1%

DISCUSSION

Note that these experiments are between processors in the same hypernode. Similar tests across hypernodes would be needed to fully characterize this architecture's point-to-point message-passing performance.
The experiments using multiple iterations show the effect of the improved memory locality intrinsic to the iteration logic, giving better performance for certain message sizes. This is unlikely to be indicative of performance in real applications.

Performance Studies using

PSTSWM

HP/Convex SPP-2000 SWAP Performance

(unordered swap of 8KB message using MPI)

DISCUSSION

Patrick H. Worley / ( worleyph@ornl.gov) Last Modified Monday, 15-Jul-2002 10:22:48 EDT. 86918 accesses since 1/2/96.

Patrick H. Worley / ( worleyph@ornl.gov)
Last Modified Monday, 15-Jul-2002 10:22:48 EDT.
86918 accesses since 1/2/96.