PSTSWM SPP-2000 Point-to-Point Communication Performance

Performance Studies using

PSTSWM

HP/Convex SPP-2000 SWAP Performance

(unordered swap of 2MB 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:	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:	(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. 322.24 75.51 67.5%

10 iter. 363.99 33.85 76.5%

unordered swap using nonblocking send

Data Statistics
bidirectional bandwidth estimated latency model error

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

1 iter. 646.96 38.02 48.1%

10 iter. 856.82 35.29 59.0%

1 iter. w/overlap 641.92 37.19 48.3%

10 iter. w/overlap 851.77 34.19 59.2%

unordered swap using nonblocking receive

Data Statistics
bidirectional bandwidth estimated latency model error

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

1 iter. 653.01 35.44 48.9%

10 iter. 821.93 32.45 58.4%

1 iter. w/overlap 647.37 38.18 47.7%

10 iter. w/overlap 858.26 33.79 60.2%

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. 654.02 35.49 49.1%

10 iter. 847.95 31.81 59.5%

1 iter. w/overlap 645.57 45.94 62.1%

10 iter. w/overlap 856.17 36.83 60.4%

unordered swap using ready send

Data Statistics
bidirectional bandwidth estimated latency model error

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

1 iter. 647.97 53.58 45.0%

10 iter. 854.12 51.13 55.7%

1 iter. w/overlap 645.68 36.08 48.2%

10 iter. w/overlap 853.21 31.94 60.4%

unordered swap using nonblocking ready send

Data Statistics
bidirectional bandwidth estimated latency model error

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

1 iter. 650.08 53.23 45.3%

10 iter. 852.14 51.03 55.0%

1 iter. w/overlap 633.29 46.14 58.0%

10 iter. w/overlap 857.31 34.87 61.4%

native sendrecv

Data Statistics
bidirectional bandwidth estimated latency model error

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

1 iter. 653.94 38.73 48.1%

10 iter. 857.22 35.60 58.5%

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 significantly better performance for certain message sizes. This is unlikely to be indicative of performance in real applications.
A change in protocol, due to changing message size, appears to be evident in the graphs.

Performance Studies using

PSTSWM

HP/Convex SPP-2000 SWAP Performance

(unordered swap of 2MB message using MPI)

DISCUSSION

Patrick H. Worley / ( worleyph@ornl.gov) Last Modified Monday, 15-Jul-2002 10:22:46 EDT. 86700 accesses since 1/2/96.

Patrick H. Worley / ( worleyph@ornl.gov)
Last Modified Monday, 15-Jul-2002 10:22:46 EDT.
86700 accesses since 1/2/96.