please take to heart so we can get real benefits of multicast at the rather strained interconnection points -- k Forwarded message:
From nobody Mon Feb 12 13:52:53 1996 X-Mailer: exmh version 1.6.4 10/10/95 To: Peter Lothberg <roll@stupi.se> Cc: Phil Dykstra <phil@arl.mil>, mbone@ISI.EDU Subject: Re: FIX/NAP Connections In-Reply-To: Your message of "Sat, 10 Feb 1996 21:40:06 PST." <CMM.0.90.0.824017223.roll@Junk.Stupi.SE> Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii Date: Mon, 12 Feb 1996 11:54:12 PST From: Bill Fenner <fenner@parc.xerox.com> Message-Id: <96Feb12.115421pst.177478@crevenia.parc.xerox.com>
I would like to push for native peering rather than tunnels wherever possible. Native peering allows the Ethernet/FDDI/whatever-physical-medium to do your packet replication for you, so instead of 10 tunnels to 10 providers meaning 10 copies of each packet, you can have a single interface on a shared medium and have only one copy of each packet. Now, this has its own expenses; if your existing routers don't support multicast (or don't support it well) then you need to put another two-interface box at the NAP & give it an interface on your router like Peter was describing. My vision of how things might work is something like T3 to provider A _____________ ________/___ | A mrouter | -------- | A router | ---- Gigaswitch port ------------- ------------ / \ T1 to provider B / \ T3 to provider C ____/____ _____/____ ____\_____ _______/___ |B Router| -- |B Mrouter| |C Mrouter| ------- | C router | --- Gigaswitch port --------- -------\--- --/-------- ------------ \______/ (I hate drawing ASCII diagrams). Basically, there is a single FDDI ring (well, it could even be an ethernet, given the quantity of multicast traffic we are currently seeing) where each mrouter has an interface and everyone peers for multicast traffic. Then each of these multicast routers has a private interface with the provider's unicast router in order to carry tunnels into the provider; this could also be an ethernet, because this box should only have one tunnel to the provider's internal multicast topology. There is no real reason for the seperate subnet, other than isolation of resources, and assuming that the gigaswitch handles multicast properly. This kind of thing is happening at MAE-East; PSI, Digex, Sprint and Sura are peering natively. I would like to see it happen at every provider interconnect. Bill