Re: Protocol Benchmarking (with Accept examples - long)

• Mail folder: WWW Talk Jan 94-present
• Next message: ellson@hotsand.att.com: "Motif 1.2.2 lock up problems."
• Previous message: frans van hoesel: "Re: Protocol Benchmarking (shortened :-)"
• Maybe in reply to: Lou Montulli: "Re: Protocol Benchmarking (with Accept examples - long)"
• Reply: Larry Masinter: "Re: Protocol Benchmarking (with Accept examples - long)"

Message-id: <9402030140.AA13393@dxmint.cern.ch>
Date: Wed, 2 Feb 94 20:39:21 EST
From: hgs@research.att.com (Henning G. Schulzrinne)
To: www-talk@www0.cern.ch
Subject: Re: Protocol Benchmarking (with Accept examples - long)
Content-Length: 864

To add some very basic performance numbers:
- each cross-country round-trip will take a delay of 60 ms or more
  (example: ping times from here on the East coast to the West coast
  from and to well-connected sites are on the order of 100 ms).

- a 1 KByte packet will take 8 ms of transmission time on each
  1 Mb/s (T1) hop, without any queueing. Take an example:

  ping (round-trip) time from portal.research.att.com to beta.xerox.com:
  56 bytes:   about 100 ms
  1000 bytes: about 190 ms

The arithmetic is thus fairly straightforward: if you need more than
two even tiny packets (on average) to do the job of one monster
packet, you lose. The calculation gets worse with high-latency
firewalls or transatlantic connections in the middle. For slow
links, clearly, things look somewhat different; there is no
universally best strategy. 

Henning Schulzrinne