Skip to Content.
Sympa Menu

ndt-users - Re: Spurious results

Subject: ndt-users list created

List archive

Re: Spurious results


Chronological Thread 
  • From: Richard Carlson <>
  • To: "chet" <>, <>
  • Subject: Re: Spurious results
  • Date: Thu, 27 Oct 2005 16:27:28 -0400

Hi Martyn;

Comments in-line
At 03:11 PM 10/27/2005, chet wrote:
Is there an analyser I can use to read the results displayed by ndt, I'm getting weird results when running tests under different environments, back-to-back tests produce results similar to: -

client to server 84.86Mb/s
server to client 62.86Mb/s

Different results for different clients or after moving a client to a new location are to be expected. That's one of the reasons I started this project, I couldn't easily document all these changes.

I'm guessing I have got to do the /etc/sysctl.conf edit to resolve the above issues but need to ask first

If I put a HUB in the way then these results drop by around 20Mb/s, these are not the exact results as I'm not in the office now

The HUB is a shared device, so the hosts will drop to half-duplex. Half duplex connections will run at 60 - 70 % of the link capacity. This is due to the Ethernet CDMA-CA protocol and should be expected. (I'll note here that the half/full detection algorithm probably doesn't work right now so you'll get a report that the path is full duplex).


My set up for back-to back results are both server and PC set to 100 half duplex, server has twin p3 500 with 512 meg of ram, my laptop is a Pentium p4 with 256 MB of ram

If I go across a VLAN the Cat 5509 has it ports set to 100 half duplex and the Cat 2900 is set to 100 half duplex, if I set them all to 100 full duplex I get the following: -

Checking for Middleboxes . . . . . . . . . . . . . . . . . . Done
running 10s outbound test (client to server) . . . . . Server busy: Please wait 30 seconds for previous test to finish

This indicates that the client can't reach the server. Check the IPTABLES to ensure that hosts on this VLAN can access the server. Or check for other firewalls that are blocking port 3002 (3003 is getting through because the middlebox test is working). Also see below for some debugging steps.


What is the optimum set up for server, is it 100 half duplex

I'd suggest connecting the server directly to a switch and running it in full-duplex mode. This will prevent the NDT server link from becoming a bottleneck, making it easier to debug other problems.

What is new in 3.1.4b as I'm using 4a, if I update to 4b I guess I leave 4a where it is and just start ndt from 4b

version 3.1.4b simply adds in the appropriate shutdown() commands into the server to prevent multiple clients from interfering with each other. If you have patched the web100srv.c file by hand, you do not need to download/install the 4b version.

Lastly, the NDT server supports a wide variety of debugging and analysis functions. Most of them are enabled by command line options. The interesting ones are:

-t This option causes the NDT server to write a tcpdump file for every
test. These files can be viewed via the tcpdump program or you can]
create graphical plots using the tcptrace program. (search the web if
you want more details on these programs). One simple debugging
task is to run the web100srv process with the -t option and verify that
2 -non zero length ndttrace-ip.addr.port.number files are created. This
indicates that the test ran and some data was collected.

-x This option causes the NDT server to run any experimental code. At
the present time the this will cause the server to write a Web100
snap-log file for each server-to-client test. You can then use the
Web100 watchsender program to view the collected data. Not much
you can do with this, but it is available

-d This option turns on debugging. The more -d flags you specify the
more debugging details you get. Messages are written to the stderr
interface so you need to redirect them to a file or run the web100srv
process in the foreground.

Finally, there is a small program /usr/local/bin/analysis that parses the
web100srv.log file and writes out a bunch of details. You may find this
useful or not.

Regards;
Rich
Rich

Is there a analyser that I can use to manipulate the results

Thanks

Martyn

------------------------------------



Richard A. Carlson e-mail:

Network Engineer phone: (734) 352-7043
Internet2 fax: (734) 913-4255
1000 Oakbrook Dr; Suite 300
Ann Arbor, MI 48104



Archive powered by MHonArc 2.6.16.

Top of Page