[5356] in www-talk@info.cern.ch
Re: Virtual Circuit protocols => universal access
daemon@ATHENA.MIT.EDU (Karl Auerbach)
Fri Aug 26 03:29:06 1994
Date: Thu, 25 Aug 1994 18:05:52 +0200
Errors-To: listmaster@www0.cern.ch
Errors-To: listmaster@www0.cern.ch
Reply-To: karl@cavebear.com
From: Karl Auerbach <karl@cavebear.com>
To: Multiple recipients of list <www-talk@www0.cern.ch>
> >What about a reliable, connectionless protocol, like ARDP? (Well, it's
> >not really any more or less connectionless than TCP, but there is no
> >connection establishment overhead in ARDP).
>
> Maybe the connection overhead concerns can be solved another way such
> as with MGET, lookahead, smarter caching etc. ?
>
> On the otherhand, it might be reasonable to implement HEAD with a datagram
> if that would help.
(This kind of stuff tends to be discussed on the www-speed list.)
That might help a little bit, but it means putting timers and
retransmit logic in the client.
But my observations indicate that transfer delay and rate are governed
quite a bit more by TCP connection time and the progression through
its slow start algorithms.
FYI...
Just a data point -- two DEC Alpha boxes (i.e. relatively fast
engines), one with a the NCSA server and the other with mosaic. They
each have short, direct 155Mbit paths to a dedicated ATM switch.
In other words, this setup has significant CPU and bandwith resources.
And the time to fetch documents wasn't significantly faster than on an
Ethernet.
--karl--
