Re: [Tomahawk-users] Trouble reproducing CPS results

[Koconis, D. <dav...@ic...>]

Andrey,

Although I did not participate in the testing cited by the whitepaper, I
have contacted one of the authors who provided me with the cps script
referred to on page 20 and gave me permission to share it with you and
the list.  It is attached to this email.

>From what you have posted about your configuration, I do not see any
issues.  I also reviewed the 1000conn.pcap you created and it looks very
similar to the one the author provided me.

One important thing to know before attempting to run the script is it
will only work with tomahawk version 1.0.  Some of the changes made in
release 1.1 will cause the command line in the script to fail.
Specifically, the maximum number of handlers in v1.1 is 250, so "-n
10000" will fail.  Also, the second byte of the IP address, which is a
shell variable $ADDR in the script, is now reserved for use as the
handler ID.  Although I have not verified it, I believe the following
changes should make things work with v1.1:

	loops=3D1000
	time tomahawk -i eth0 -j eth1 -q -m 200 -r 10 -t 100 -a 10.0.0.0
-n 250 \
		-l $loops -f /usr/local/pcaps/1000conn.pcap > /dev/null

If you are still having difficulty getting the results you want, please
post the command line you are using and the version of the code.

Regards,
David Koconis

-----Original Message-----
From: tom...@li...
[mailto:tom...@li...] On Behalf Of
Andrey Falko
Sent: Monday, October 01, 2007 10:12 PM
To: tom...@li...
Subject: [Tomahawk-users] Trouble reproducing CPS results

Hi everyone,

I hope someone still reads this list. We are trying to achieve the
connections per second recorded in the white paper (76,500 cps) on page
20. We are getting only around 9865 cps.

We are using the cps script provided in qa.tgz. We had to generate our
own 1000conn.pcap trace. We generated 1000 TCP connections with full
3-way handshakes using a Perl script that that connected to an apache
server on a machine on the network (I have attached the tcpdump of that
transaction and the Perl script). Is this trace proper?
Is it possible to get the trace the author utilized in the white paper?
Are we supposed to be generating full 3-way handshakes another way?

Our hardware consists of six machines with the following specs:

AMD64 4400+ CPU
2G of DDR memory
Two Intel NICs (eth0 and eth1)
One Nvidia NIC (eth2, which is the management port)

These machines are connected to GigE 48-port and 24-port Netgear
switches. The Intel NICs are one the same subnet.

We also tried to utilize one dual AMD64 machine with 12G of ram and two
Broadcom GigE adapters connected to a Dell GigE switch. Our results were
the same.

Thank you in advance for any suggestions and help.

Best regards,
Andrey Falko