Hello all,

This is my first post, but I've been going through this list a lot, and it has really helped me in many occasions.

I would like to come up with a script to test the network performance between:

1. 2 Sun clusters
2. 1 Sun cluster and a windows workstation

I don't want to involve any third party programs. I just want to use programs local to both systems. Pings, tracerts, FTPs are OK.

I want something that can tell me the thoroughput. I want to use it as a benchmarking tool for determining TCP and UDP performance between the 2 systems.

I would also like to know the procedure for raising network limits under Solaris. I'm aware that the ndd variables (tcp_max_buf, tcp_xmit_hiwat, tcp_recv_hiwat) can be set, but how?

Please help.

Thank you.

If your goal is only about the throughput, standard ftp is already giving you this measure after each transfer:
ex: 1148932 bytes received in 0.13 seconds (8944.62 Kbytes/s)
The numbers show the workload, the payload is always higher.

For udp, perhaps a timed nfs copy will do the job.

Of course, watch your disk load while testing to be sure you really measure network and not disk performance. To avoid the issue, the simplest way is to work on /tmp (tmpfs), and assure the system has enough RAM (no paging).

About changing network driver's variables, an example of the syntax to use is:

# ndd -set /dev/tcp tcp_xmit_hiwat 65536

Beware that you can easily degrade vs improve a system by wrongly "tuning" it.

Thank you for your prompt response.

I'm using this ftp script for the throughput measurement:

#!/bin/bash

HOSTNAME= xxx.xxx.xxx.xxx

ftp $HOSTNAME << !
bin
pipe
put "| dd if=/dev/zero bs=32k count=10000" /dev/null
bye
!

It helps. I have'nt been able to test the timed nfs copy yet.

I think I've seen a shell script out there somewhere that does what I want, but I can't seem to remember where.

Any more help would be appreciated.

Thank you, again.

For nfs, I don't think the trick of using /dev/null or /dev/zero as a remote file would work, so if you want to avoid disk interference, what you can do is sharing tmpfs directories on the remote host like this:

share -F nfs -o rw -d "/tmp" /tmp

then:

time dd if=/dev/zero bs=32k count=10000 of=/net/remote/tmp/320MB

will give you the UDP time, as long as you start your nfs server with the option to lock it in UDP (-p udp i think, or in /etc/default/nfs if this file exists, on solaris 9 maybe and 10 for sure)

You can also use tftp for udp benchmarks.

Thank you, once again.

I'm going to try your method. It looks fine to me, and I hope to have the results of my tests soon.

I appreciate your help.

Thank you.

I would suggest nsar it is available at http://www.sunfreeware.com . You can go all out and have mrtg going as well http://people.ee.ethz.ch/~oetiker/webtools/mrtg/

Thank you, neo77777.

I'm always looking for better tools or scripts that might help me with the original task.

I'll be checking out the tools that you've mentioned. However, mrtg might be an overkill.

Thank you, and I appreciate you help.