Re: Networking stalls: More details
Michael Neuffer (neuffer@goofy.zdv.uni-mainz.de)
Mon, 19 Aug 1996 12:35:01 +0200 (MET DST)
On Sat, 17 Aug 1996, Tor Arntsen wrote:
> Alan Cox <alan@cymru.net> writes:
> [someone else writes:]
> >> (and you'll hit the maximum very fast on a network with 50% packet
> >> loss). This has to be fixed (trivial, fortunately).
> >
> >Its not trivial without breaking the basic algorithms. Ultimately the
> >problem you have is that a transmit/ack based protocol is not a good
> >answer to that much loss. You either need something other than TCP on
> >top or to fix the link layer in the case of amateur radio links.
>
> Actually, the Internet (at least in Europe, and particularly for us C-domains
> which get treated really bad in the routers -- according to our ISP) is now
> so overloaded that packet loss is a very real problem also without any radios
> or phone lines involved..
> I'm doing remote maintenance work over the Internet (only using telnet and
> ftp), and it's now close to impossible. Packets get lost, and tcpdump tells
> me it's longer and longer between acks, I press <return>, I get myself a
> cup of coffee, absolutely nothing happens, more coffee..
You're taking the words out of my mouth. Due to this, maintenace work
across the atlantic is often nearly impossible.
Here packet loss rates for connections accross the atlantic are normaly in
the 30%-65% range.
Last sunday morning around 4am the connection was exceptionally good:
only about 16% packet loss....
Mike