[e2e] packet loss in the Internet

Jonathan Stone jonathan at DSG.Stanford.EDU
Fri May 24 11:43:09 PDT 2002


NWAY itself works fairly well.  One problem is that vendors sold a fair
number of (then) high-end 10/100 switches (cisco catalyst 2900,
first-gen 100mbit blades for Catalyst 5000) which predated
NWAY standardization (or perhaps patent issues with NatSemi).

Partly for this reason, some network admins prefer to disable the
autonegotiation protocol and hand-configure full-duplex.
Subsequent changes by another admin who doesn't follow the same
convention often leads to half-duplex/full-duplex mismatches.
Similar problems can arise at the edge of the enterprise networks,
with old 10/100 hubs, which are half-duplex-only devices.

My experience is that the problem happens often enough that
co-loc engineers are quick to spot it.

In message <000d01c202d2$47e23530$05a8a8c0 at thehole>,
"Stefan Savage" writes:

>I agree about congestion.  However, I've measured significant loss in
>enterprise networks on several occasions that arose from Ethernet
>full-duplex/half-duplex conflicts (its still not clear to me if the
>Ethernet autonegotiation protocol is fundamentally inadequate or if
>current implementations are simply poor).  Its unclear to me how
>prevalent this problem is, but I've run into it in three separate
>- Stefan

More information about the end2end-interest mailing list