ndt-users list created

[Richard Carlson <>]

Jordi;

At 02:58 PM 4/23/2008, Jordi Bosch wrote:
> Hi All,
> We are considering running NDT as a bandwidth test tool. Maybe 
> Iperf/BWCTL would be better, anyone can tell me some public servers 
> to see how it looks like?

See
http://e2epi.internet2.edu/ndt/ndt-server-list.html and 
http://e2epi.internet2.edu/pipes/pmp/pmp-dir.html for links to public 
NDT and BWCTL servers.

> I've observed looking at the 'network tools' that each NDT test 
> sends 10 MB of data in the S2C test, and about 1,3 MB in the C2S 
> test.I've looked at the 'Bytes sent' and 'Bytes Received' counters 
> under the network device, eth0 in this case.

The NDT system runs 10 sec tests in each direction.  The amount of 
data transferred is dependant on the how fast the hosts can move 
data.  Iperf/BWCTL sends data for x seconds, where x is a run time 
option.  There are other tools (i.e., ttcp) where a fixed amount of 
data is sent.

> From some documents i've understood that only some pairs of packets 
> are sent, but almost the Tcpbw100.java seems to send 8192 byte 
> buffers for 10 seconds. I deduced that sending only some pairs of 
> packet allows NDT overcome the TCP windows limits to get high 
> bandwith. Please, anyone can tell where i'm wrong?

The NDT sending process (C2S or S2C) runs in a tight 'do loop" 
dumping 8KB buffers down to the kernel for delivery over the 
network.  I am currently testing a change in the C2S test to move 64 
KB blocks down to the kernel to overcome a problem with Windows based 
hosts.  The kernel is responsible for breaking these buffers into 
packets and TCP then delivers them across the network.  All the data 
generated by the application should be sent to the remote receiving 
host.  I would be interested in seeing network traces that show some 
data blocks getting dropped by the kernel.

> I'm concerned about TCP window parameter, wich in fact can limit the 
> size of the pipe that can be measured.I would like to ask about the 
> NDT ability to test up to 1 Gbps pipes: can NDT test 1 Gbps pipes 
> despite of TCP window limits ?

Yes, TCP window parameters can, and do, effect the amount of data 
that the NDT sending host can push into the network.  One benefit of 
the NDT system is that the web100 kernel reports this parameter so 
the S2C test will report if the remote client is so limited.  This 
makes it easy to understand why the throughput was less than expected.

To answer you're underlying question, the NDT system has no problems 
running at multiGigabit rates over 10 Gbps Ethernet NIC's.  In the 
Internet2 national backbone, with properly configured and tuned 
hosts, we can achieve 900+ Mbps over transcontantial paths.

Rich
> Thanks in advance for your time, and of course thank for this great tool!
> Regards.
> Jordi.

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



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