[e2e] Unexpected small reduction in cwnd during iperf test
toby.rodwell at dante.org.uk
Tue Aug 8 15:45:07 PDT 2006
Many thanks for your reply.
Xiaoliang (David) Wei wrote:
> Hi Toby,
>> ... The TCP transfer in question was an iperf test
>> running between a measurement point in a New York PoP and
>> another in Budapest. netstat (run before and after the test) confirms
>> that there was no packet loss, but there were two re-transmitted
> I wonder how netstat could tell if there is packet loss? You mean "no
> packet loss at the local NIC?"
"netstat -s" on the sender shows total number of TCP segments lost -
since there was no difference between the value before or after the test
I'm confident there was no packet loss.
> as the cwnd before this glitch has been kept to 5931 unchanged for
> several RTTs (I assume the RTT in your case is in the order of 100ms
> across the Atlantic), I guess one possibility is that there is some ack
> packet reordering.
It is certainly possible (almost likely) that ACKs are being reordered,
but I hadn't realised that would have any knock on effect - I'd always
thought that a disordered 'late' ACK would be ignored as the disordered
'early' ACK would have acknowledged its sequence space.
> just one explanation... not sure if it is of your case. More
> measurements on the TCP congestion avoidance state is helpful, I think.
> Or, if the experiment is repeatable and you could have the tcpdump trace
> during this events, it is also helpful.
The experiment is certainly repeatable but the end hosts are not
particularly powerful, and I don't think they would support trying to
collect tcpdump traces (even just the headers) at 650Mbps without
adversely affecting the result.
DANTE - www.dante.net
Tel: +44 (0)1223 371 300
Fax: +44 (0)1223 371 371
Email: toby.rodwell at dante.org.uk
PGP Key ID: 0xB864694E
City House, 126-130 Hills Road
Cambridge CB2 1PQ
More information about the end2end-interest