PSTSWM T3D Point-to-Point Communication Performance

Performance Studies using

PSTSWM

Cray Research T3D SWAP Performance

(ordered swap of 2MB message using SHMEM)

Date/Person:	November 3, 1994 / P. Worley
Platform:	T3D at Cray Research in Eagen, MN (typhoon):
	512 150-MHz DEC Alpha EV4 processors
Environment:	???
Communication Library:	SHMEM
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 2
Latency Definition:	(T₁₀₂₄-T₅₁₂)/512
Results:

ordered swap using nonblocking send

Data Statistics
unidirectional bandwidth estimated latency model error

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

1 iter. 36.87 18.98 0.2%

10 iter. 36.87 18.87 0.1%

1 iter. w/overlap 36.98 18.88 0.4%

10 iter. w/overlap 36.97 14.42 0.2%

ordered swap using nonblocking receive

Data Statistics
unidirectional bandwidth estimated latency model error

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

1 iter. 126.36 11.79 5.7%

10 iter. 126.36 11.81 5.8%

1 iter. w/overlap 126.36 19.70 1.1%

10 iter. w/overlap 126.31 16.16 1.3%

DISCUSSION

A T3D processing node consists of two processors that share the node's network interface hardware. When exchanging data, it can be slower between processors in the same node than between processors on different nodes, due to contention for shared resources. For this reason, our SWAP experiments used logical processors 0 and 2 in a 4 processor partition.

Performance Studies using

PSTSWM

Cray Research T3D SWAP Performance

(ordered swap of 2MB message using SHMEM)

DISCUSSION

Patrick H. Worley / ( worleyph@ornl.gov) Last Modified Monday, 15-Jul-2002 10:24:09 EDT. 86406 accesses since 1/2/96.

Patrick H. Worley / ( worleyph@ornl.gov)
Last Modified Monday, 15-Jul-2002 10:24:09 EDT.
86406 accesses since 1/2/96.