Re: about iperf
On Tue, 15 Aug 2006 00:32:13 -0400
Matthew J Zekauskas <matt --at-- internet2.edu> wrote:
> One thought -- this sounds like a classic MTU black hole. The
> negotiations are done with small packets, but when the first big packet
> goes out, it is dropped and the connection hangs. Maybe a VLAN or
> something wasn't set up properly. Is the MTU the same for all servers?
>
> You could test this by having a ssh connection from c to b (or b to c),
> and then try to cat or otherwise print a large file to the terminal. If
> it's an MTU black hole (and TCP for ssh negotiates the same MSS as with
> iperf I suppose), that connection will also hang.
>
> It could also be some security/firewall thing, too. You could also try
> a tcpdump on either side, and see what is going on.
>
> --Matt
>
>
> On 8/14/2006 11:37 PM, Bob wrote:
>
> > Hi, all,
> >
> > I am currently testing a link. I have 3 servers located in three points separately. The 3 servers are all running linux OS.
> >
> > The topology looks like this:
> >
> >
> >
> > Point1 Point2 Point3
> >
> > A-------------------------------------------------------ï
> >
> > ---------------------------------------------C
> >
> > B__â
> >
> > I can traceroute between A-B, B-C, A-C successfully. And I can run iperf between A-B, A-C.
> >
> > But when I ran iperf between B-C, no matter I configure B as server or client, iperf will halt after connection is built. I donât know what the reason is. It seems that network connections between the three points are ok. Could this problem be caused by security configuration?
> >
> > Have you ever met such problem? Please give me some suggestions.
> >
> > Thanks.
> >
> > Bob
>
Also, check for protocol corrupting middleboxes that don't do TCP window scaling correctly.
Some security products (OpenBSD firewall seems to be common), don't correctly remember the window
scale and cause the connection to die. Unfortunately, it is not something that Linux can
correct for automatically. You can turn off window scaling but that limits performance.