I did not work *for* UUNET either. However, I did work for Cascade *at* UUNET as on on-site consultant sometime in 1995. I worked with their network engineers in analyzing traffic patterns, performing network design, and looking at network costs. In addition, I followed their network design closely from 1994-1998. I can definitely state that UUNET had its own separate Frame Relay and (later) ATM backbone network that they did not share with Worldcom's Frame Relay/ATM network. UUNET's L2 switches were directly connected to T3 TDM and later OC-12 TDM links. Prabhu

-----Original Message----- From: Craig Partridge [mailto:craig@aland.bbn.com] Sent: Wednesday, August 08, 2001 9:19 AM To: Kavi, Prabhu Cc: nanog@merit.edu Subject: Re: MPLS VPNs or not?

In message <6B190B34070BD411ACA000B0D0214E56CB8128@newman.tenornet.com>, "Kavi, Prabhu" writes:

...

I don't think UUNET considered it a waste. UUNET could not have grown as quickly as it did during the mid to late 90s without L2 (Frame and ATM) technologies. Fortunately for them, they did not have any pure IP only zealots that prevented the pragmatic use of other technologies in their networks. Otherwise they probably would not have been able to outrun the other ISPs.

I know the history here pretty well (though since I was never employed by UUNET, I'll probably get some of it wrong). It doesn't quite match this description.

UUNET was an excellently run ISP well before it starting doing the ATM activity. It came (I suspect) from Rick Adams' days as a struggling ISP, competing against Govt subsidized regional networks, in the 1980s. If you look at their various SEC filings in their first few years, they're a tightly run company.

Once UUNET was acquired by Worldcom, UUNET had access to Worldcom's network infrastructure, which was heavily ATM, and which UUNET had to share with the Worldcom's (very lucractive) voice traffic. In that context, there was a real cost to using bandwidth and UUNET had to use account for its usage.

Some other ISPs were/are in a different business model -- they owned their fiber runs, outright, and the question for them was whether to put run ATM over that fiber, and subdivide the bandwidth of a single waveband, or light two wavebands (one voice/one IP). I've seen the marginal cost analysis for that kind of decision, and it often favors two wavebands.

Craig