ndt-users list created

[Richard Carlson <>]

Ralf;

One comment on the bottleneck link detection.  Then NDT server needs to 
process every packet as it arrives.  Most Gigabit NICs default to using 
interrupt coalescing schemes to reduce the CPU overhead.  You must turn 
this feature off on the NDT server so the ndtd process sees every packet 
when it arrives.  The ethtool utility can be used to turn off this feature.

sudo ethtool -c eth0  (show the coalescing values for eth0)
sudo ethtool -C rx-frames 1 eth0 (disable coalescing on eth0)

There should be something in the readme files that describes this, if 
not it needs to be added.

Rich

On 1/14/2010 9:04 AM, Ralf Hildebrandt wrote:
> * Aaron 
> Brown<>:
> > On Jan 14, 2010, at 7:52 AM, Ralf Hildebrandt wrote:
> >
> > > * Jason 
> > > Zurawski<>:
> > > > All;
> > > >
> > > > Version 3.5.14 of NDT is now available.  This release should address
> > > > the queuing problem that was reported earlier.
> >
> > You were hitting on a bug in a library that NDT depended on. We've
> > uploaded a new version of the 3.5.14 package (new md5 checksum:
> > d1dc147018f155f96674e8c5e0e9b8ad ). Let us know if this doesn't fix the
> > issue.
>
> Seems to work now.
>
> TCP/Web100 Network Diagnostic Tool v3.5.14
> click START to begin
>
> ** Starting test 1 of 1 **
> Connecting to 'rispacs-perf.charite.de'
> [rispacs-perf.charite.de/10.48.4.4] to run test
> Connected to: rispacs-perf.charite.de  --  Using IPv4 address
> Checking for Middleboxes . . . . . . . . . . . . . . . . . .  Done
> checking for firewalls . . . . . . . . . . . . . . . . . . .  Done
> running 10s outbound test (client-to-server [C2S]) . . . . . 914.98Mb/s
> running 10s inbound test (server-to-client [S2C]) . . . . . . 888.41Mb/s
> The slowest link in the end-to-end path is a a 622 Mbps OC-12 subnet
> Information: Other network traffic is congesting the link
>
> After that the status stays at "Receiving results..." and all controls
> stay grayed out.
>
> Note that the bogus "slowest link" message is still there