Average and median IP packet size over the Internet?

Tue Apr 28 03:13:05 UTC 1998

On Mon, Apr 27, 1998 at 04:44:15PM -0700, Michael Dillon wrote:
  On Mon, 27 Apr 1998 Goldstein_William at bns.att.com wrote:

  > Anyone know of any studies documenting this?  A know that IP packets
  > over the Internet tend to be smaller rather than larger, but I wondered
  > if anyone has published something.

  You are likely to find that sort of thing here http://www.caida.org/ 

right now it's at 
http://www.nlanr.net/NA/Learn/packetsizes.html
(pointed to from
	http://www.nlanr.net/NA/tutorial.html
specifically 
	http://www.nlanr.net/NA/Learn/plen.970625.gif
from 1997

note difference if you look at in terms of packets vs bytes
(sorry to those who have heard this a million times;
am trying to only say it once a week whether need to or not):
more than half the *packets* are mice,
more than half the *payload* are elephants

[note that CAIDA hopes to update this whole tutorial thing
by summer (some data quite old), 
willing hands/brains cooperating
so don't holler at me for archaic questions/answers,
but now is the time to provide suggestions on things you'd like
included/extended/cut, don't wait till you're disappointed]

but packet sizes aren't changing that fast anyway;
am as we speak am submitting an Inet'98 paper with mcifolk 
that has stats not far off from above
i.e., packets getting a little larger, 
but not enough to buy/sell any stock over

also, fwiw (have beat this URL into the ground by now, 
re-sorry broken record-ness):
	http://www.vbns.net/presentations/papers/MCItraffic.ps 
has mucho cool, surprisingly hard to come by WAN backbone stats
by amazingly dedicated angels of MCI/NSF's vBNS project
(US tax dollars + MCI karma at yummy work...
tho warning TEST the postscript before you print it
(<<i speak for the trees!>>);
and it craves a color printer)

and as already demonstrated by Al Reuben 
and many others at nanog could show you,
real world routers (well, on planet Cisco anyway,
which seems to support life)
can give you this very distribution at your whim

(k learns: it's possible that all customers 
have to do is ask for stat collection louder and longer 
than they ask for other stuff
and some of it gets implemented mmmm..
just like everything else in life)

k