RE: UDP Packet Loss in Question
Thanks for the input Hans.
I looked at the IPERF sequence number of the last packet (drilled into
the packet on Wire Shark) on both sides and they match, so that might
not be an issue in 1.7.0. I do notice on both sides though that the
sender did send an extra packet with sequence number FFFF.... this might
be a "I'm finished sending, send back the results" type message, I don't
know.
I did increase the UDP buffer size on the server and client (the -w
option) to 64k and all of a sudden, my lost packet count for IPERF = my
packet counters on Wire Shark, and it was significantly lower than the
packet loss statistic below, perhaps in the order of 0.001% now.
Maybe the receiver was running out of buffer and couldn't process the
stream (only 2.5M though!) fast enough. It's a PIII 850MHz with 2G of
RAM running Win2k3 SP1.
I'll do some more testing and verify if indeed the -w parameter fixed
things and post the results.
Cheers,
Barry
-----Original Message-----
From: owner-iperf-users --at-- dast.nlanr.net
[mailto:owner-iperf-users --at-- dast.nlanr.net] On Behalf Of Hans Blom
Sent: Wednesday, May 02, 2007 8:10 PM
To: iperf-users --at-- dast.nlanr.net
Subject: Re: UDP Packet Loss in Question
Hi Barry.
In "src/Client.cpp" from Iperf V. 2.0.2 it looks like that the last
packet ID
that is reported in the output has been reorted to the server and not
what is
really the last packet ID. This could result in untrue "datagrams
out-of-order"
messages, we belief, but their number is very limited. But I do not know
if this
was also an issue at V. 1.7.0. Possible this was not the case.
Regards,
Hans
On Wed, May 02, 2007 at 04:04:41PM -0230, Carew, Barry wrote:
> Hi,
>
>
>
> I am seeing a discrepancy in packet counters on IPERF 1.7.0 and on
Wire
> Shark, which I think is packet loss numbers. First I am running IPERF
> 1.7.0 on two Win2K3 server boxes, separated by an MPLS network. I run
> IPERF client on one box and IPERF server (both in UDP mode) and send
for
> about half an hour @ 2.5M using 1400 byte packet size (the network
> allows 3M and the MTU is 1500). The conclusion of one of my tests is
> below:
>
>
>
> [1924] 0.0-1799.9 sec 536 MBytes 2.50 Mbits/sec 8.214 ms
> 61/413505 (0.015%)
>
>
>
> If I count the number of packets sent by the client and received by
the
> server using Wire Shark (with a filter to look at just UDP packets
sent
> to port 5001 at size 1400 bytes) I see 413505 packets on the sending
> side and 413502 on the receiving side.
>
>
>
> I have also set up some packet debug on the routers that these two
> servers connect to and looked at the total number of packets sent and
> received at each side - these numbers always match Wire Sharks
numbers.
> So it doesn't look like I'm actually losing the packet, is it because
> IPERF is looking at something else? I use the default buffer size of
8K
> for the server side of UDP; perhaps I should try increasing that.
>
>
>
> Has anybody seen anything similar in the past? Any comments
welcomed!!
>
>
>
> Cheers,
>
> Barry
>
>
>
> The information contained in this email is strictly confidential and
subject to the Confidentiality Notice at
http://www.stratosglobal.com/email-privacy.html
>
>
>
The information contained in this email is strictly confidential and subject to the Confidentiality Notice at http://www.stratosglobal.com/email-privacy.html