Next:
p2p_dummy
Up:
Point-to-point-pattern
Previous:
MPI_Sendrecv_replace
 
Contents
MPI_Ssend-MPI_Recv
In this measurement we use
MPI_Ssend
for sending and
MPI_Recv
for receiving. Here we can fix the overhead of synchronous sends.
Pattern: Point-to-Point varied over message length.
default values: 8 nodes, max. / act. time for suite disabled/0.75 min.
Max. allowed standard error is 3.00 %.
message length
X axis scale
logarithmic
Param. refinement
automatic x wide adaption
Argument range
1 - 8388608 units (s. below)
Argument stepwidth
1.414214
Units: bytes in type: basetype 1 (MPI_INT), send-datatype 20: size 4 Bytes, extent 4 Bytes (user defined basetype 1 (MPI_INT)), receive-datatype 20: size 4 Bytes, extent 4 Bytes (user defined basetype 1 (MPI_INT))
Latency at msg. length 0 units: 572.2
s
Bandwidth at msg. length 0 units: 0 units/s
Latency at msg. length 8192 * 1024 units: 1.51451e+06
s
Bandwidth at msg. length 8192 * 1024 units: 10818 * 1024 units/s
Per Ekman 2002-01-29