[ih] Packet nets not connected to the ARPAnet

[Guy Almes]

Dave,
   As you say, CSNet did an IP-over-X.25 service.  We had one of these 
at Rice University 1984-87.  A VAX with an X.25 card served as our 
border router.  It connected, via the 9.6kb/s GTE Telenet X.25 service, 
to some equipment at BBN in Cambridge.
   When the X.25 cloud was lightly loaded, things worked moderately 
well.  But when things got congested, let's just say that performance 
became very bad and we'd typically end up paying about $4,000 per month 
in GTE Telenet "packet charges" (totally apart from CSNet dues).
   Things got better in 1987 when our 56kb/s ARPAnet connection and a 
second 56kb/s connection to the "fuzzball" proto-NSFnet.
   And again, much much better when the T1-based NSFnet came in July 1988.

   So in 1988, when Van Jacobson kindly sent emails describing his TCP 
work, we had a massive "aha" moment, realizing that most of our $4,000 
in packet charges were probably for useless (and counterproductive) 
retransmissions.

   Regards,
	-- Guy

On 7/25/12 9:38 AM, Dave Crocker wrote:
>
> On 7/23/2012 4:18 PM, Alex McKenzie wrote:
>> CSNET (funded by the NSF) used both ARPAnet and
>> Telenet for transport but only provided user-level (eg mail)
>> interconnection and only for CSNET members.
>
>
> Initially, CSNet was strictly email transfer, but it evolved to generic
> packet services.
>
> The initial interconnection was between dial-up access to a CSNet relay
> and the Arpanet.  First use of Telenet was actually dial-up across X.29
> (telnet equivalent).  CSNet saw Telenet as a dial-up link.
>
> Its "phonenet" service layer put a packet discipline over the dial-up,
> which meant that I needed to tailor its packet size to match the one
> used by Telenet, or traffic charges doubled.  Eventually CSNet did
> provide IP over X.25.
>
> Until using IP, strictly speaking CSNet was connecting hosts to the
> Arpanet, rather than doing what we call internetworking.
>
> d/
>
>