ndt-users list created

[Richard Carlson <>]

Hi Ken;

It certainly looks like there is a problem with the path between your 
server and client.  I'd recommend looking at every link to see if there is 
a duplex mismatch somewhere.

I've removed some lines to emphasis the interesting parts.

Rich
At 07:47 AM 11/9/2006, Ken Bellars wrote:

> running 10s outbound test (client-to-server [C2S]) . . . . . 37.73Mb/s
> running 10s inbound test (server-to-client [S2C]) . . . . . . 1.07Mb/s
> The slowest link in the end-to-end path is a 100 Mbps Full duplex Fast 
> Ethernet subnet

OK asymmeteric throughput on a FastE network.  This usually indicates some 
type of non-congestive loss condition.


> Web100 reports the Round trip time = 13.26 msec; the Packet size = 1460 
> Bytes; and
> There were 123 packets retransmitted, 288 duplicate acks received, and 337 
> SACK blocks received
> The connection stalled 22 times due to packet loss
> The connection was idle 4.84 seconds (40.33%) of the time

From below, there were 1218 pkts sent and at least 10% of them were 
retransmissions.  Timeouts caused the connection to sit idle for over %40 
of the test time.  This means that no traffic was flowing during these time 
periods.


> Excessive packet loss is impacting your performance, check the 
> auto-negotiate function on your local PC and network switch

The loss rate is high so I'm suggesting you look for a duplex mismatch 
condition.  This is kind of a backdoor for me because the regular duplex 
mismatch test didn't detect a fault.


> ================ MORE DETAILS =================
> WEB100 Kernel Variables:
>
> CurCwnd: 1460
>
> CurSsthresh: 2920

SSthresh is only 2 packets,  and CWND is only 1.  Typical of duplex 
mismatch conditions.

> DataBytesOut: 1802640
> DataPktsOut: 1218
> DupAcksIn: 288
> FastRetran: 60

retransmitted 60 pkts using the fast retransmit method

> MaxCwnd: 13140
>
> MaxSsthresh: 5840

SSThresh reached a max of 4 pkts and CWND was 8.

> PktsIn: 773
> PktsOut: 1218
> PktsRetrans: 123
>
> loss: 0.067323481

6.7% packet loss


> CWND-Limited: 9467.70

limiting the CWND value to 2 pkts increases the throughput from 1 Mbps to 
9.4 Mbps.  Clearly better, but nowhere near full line rate.

So in conclusion, it looks like this path has a high non-congestive loss 
rate.  If it really is a FastE path, then looking for a duplex mismatch on 
one of the links is a worthwhile step.  If non are found, then look at the 
switch/router interface stats. I'd expect to see some errors being recorded 
somewhere.

Rich

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



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