This repository has been archived by the owner on Apr 6, 2021. It is now read-only.
-
Notifications
You must be signed in to change notification settings - Fork 0
/
README
42 lines (33 loc) · 1.71 KB
/
README
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
This is a benchmark tool for measuring UDP/TCP/MPI communication performance
in High Performance Computing (HPC) environments.
Compilation
Linux :
[~hpcbench/ @node1] % make
SunOS : uncomment the "#LIB = -lsocket -lnsl" line for each makefile in
~hpcbench/[udp/tcp/mpi] dirctories, then
[~hpcbench/ @node1] % make
Testing
1. UDP latency/throughput tests:
[~hpcbench/udp/ @node1] % ./udpserver &
[~hpcbench/udp/ @node2] % ./udptest -h node1
Refer to ~hpcbench/udp/README for more information about the UDP benchmark
2. TCP latency/throughput tests:
[~hpcbench/tcp/ @node1] % ./tcpserver &
[~hpcbench/tcp/ @node2] % ./tcptest -h node1
Refer to ~hpcbench/tcp/README for more information about the TCP benchmark
3. MPI latency/throughput tests:
[~hpcbench/mpi/ @node1] % mpirun -np 2 mpitest
Refer to ~hpcbench/MPI/README for more information about the MPI benchmark
Test scripts
There are several test scripts in ~hpcbench/testscript directory, for scaning
a set of parameters of throughput tests between two nodes. For example:
[~hpcbench/testscript/ @node1] % ./test.sh tcp node5 node6
This will examine the throughputs with different configuration (buffer size,
stream and ping-pong, blocking and nonblocking mode, etc.) between two processes,
inside both node5 and node6, and between node5 and node6. Similarly,
[~hpcbench/testscript/ @node1] % ./test.sh all node5 node6
will test all three (UDP/TCP/MPI) protocols automatically.
The scripts of test.lsf and test-all.lsf are designed for HPC's LSF queuing
system. Please refer to ~hpcbench/testscript/README for more information.
Extra tool (sys/sysmon):
Sysmon -- A Linux tool to monitor system's CPU/Memory/Network information.