Path: utzoo!utgpu!news-server.csri.toronto.edu!rpi!zaphod.mps.ohio-state.edu!cis.ohio-state.edu!ucbvax!CS.UCL.AC.UK!J.Crowcroft
From: J.Crowcroft@CS.UCL.AC.UK (Jon Crowcroft)
Newsgroups: comp.protocols.tcp-ip
Subject: Re: IP in the UK (was Re: Fingering the English)
Message-ID: <9106072007.AA01695@ucbvax.Berkeley.EDU>
Date: 6 Jun 91 10:13:33 GMT
References: <1991Jun5.012700.21291@brolga.cc.uq.oz.au>
Sender: daemon@ucbvax.BERKELEY.EDU
Distribution: world
Organization: The Internet
Lines: 41



 >I'm interested in why layering over X.25 is being used, rather than MUXing
 >out bandwidth and using HDLC cards in the routers. would stealing 48k or 64k
 >splits from the X.25 lines really be that visible? The extra protocol
 >overheads of X.25-IP barely seem worth the cost savings to me. I'd rather

geo:

To split out bandwidth by muxing would be to pre-decide how much...
and would cost some for muxes... the current choice represents
expediency and manageablity...(though for more money, we could buy
snazzy controllable muxes).

the routers have 2Mbps lines to the backbone x.25 switches - we can
tune the amount shared twixt native x.25 and IP ... by varying clocks
etc...

for well engineered (read giant window & packet size) x.25 at 2Mbps,
you dont get x.25 window blocking IP, so minimal overhead and neet
enforcement of share... experience in other nets of IP on x.25 usually
had poor x.25 implementations (read window 2, max packet size 128),
and so negative comments from elsewhere are often (not always)
discounted (though less by me:-).

of course, I would prefer frame relay to X.25 full protocol, then
there wouldnt be weird interactions between retransmits at x.25 and
above IP, but we await this...in any case, since the backbone cable
plant has very low error rate, we see this rarely...more often is
interaction twixt congestion control inside the switch cloud, and
congestion avoidance twixt TCP end points...

Another path to IP involves the x.25 backbone switches running the IP
tunnel (and later migrating to run this over frame relay internally)
this is being tried/evaluated...as will be isdn & SMDS...

 jon
p.s. if you run an AFS filesystem that we can access, please let us
know - we are currently seeing the usual set of places
(transarc/cmu/mit etc etc) but would find it amusing to go even
further afield like as far as OZ:-)