[e2e] packet loss in the Internet

Guy T Almes almes at internet2.edu
Fri May 24 09:18:38 PDT 2002


Stefan,
  I suspect you're exactly right.
  Non-congestive causes of packet loss are one of our big current plagues, 
and the duplex-mismatch problem is an embarrassing/frustrating example of 
one specific cause.
        -- Guy

--On Thursday, May 23, 2002 20:22:44 -0700 Stefan Savage 
<savage at cs.ucsd.edu> wrote:

> 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
> networks.
>
> - Stefan
>
> -----Original Message-----
> From: end2end-interest-admin at postel.org
> [mailto:end2end-interest-admin at postel.org] On Behalf Of David P. Reed
> Sent: Thursday, May 23, 2002 3:17 AM
> To: Joe Cao; end2end-interest at postel.org
> Subject: Re: [e2e] packet loss in the Internet
>
> Typical is the wrong question for enterprise nets.   As Andrew Odlyzko
> has
> noted, enterprise nets are very lightly loaded, and are managed to be
> so,
> except when there are odd constraints or lack of management attention.
> And
> loss rates are *really* low, because congestion is low.   The more
> likely
> congestion problems in enterprise nets is at the endpoints (servers and
> access links).   I observe this in real demos all the time these days,
> where I can demo a particular app that uses uncorrected UDP for voice
> streams (with buffering to eliminate jitter) and can go for hours
> without a
> lost packet.
>
> Can't speak for ISP
> At 06:14 PM 5/22/2002 -0700, Joe Cao wrote:
>> Hi Folks,
>>
>> Does anyone know in the current real world what a typical packet loss
>> rate is in an enterprise network and on an ISP trunk?  Also if drop
>> occurs, consecutive packets might be discarded.  What is the typical
>> burst
>> size that gets dropped if RED is not enabled?  Are there any such
>> statistics available?  Could anyone get me some pointers that can help
>> me
>> find those information?
>>
>>
>> Thanks!
>>
>> Joe
>
>
>
>





More information about the end2end-interest mailing list