Re: IP over SONET considered harmful?
(b) talk to your favorite vendor(s), and ask the vendor(s) to put a "knob" that would decrement IP TTL by 1 on egress (rather than copying Tag TTL into IP TTL).
Yakov.
Yakov; As I recall, the world didn't end when the NSFNET NSSs were installed and they decremented TTL twice. Weren't they changed to decrement once, appearing as one instead of two routers? (For those who don't remember the NSFNET "routers" they consisted of an IBM RS6000 attached to each of several T1/T3 interfaces, interconnected across a FDDI with a route server. Nine RS6000s in all. 200 amp service required. :-) This is no reason to raise a harmful alarm. --Kent
At 02:57 PM 3/20/98 -0800, Kent W. England wrote:
As I recall, the world didn't end when the NSFNET NSSs were installed and they decremented TTL twice. Weren't they changed to decrement once, appearing as one instead of two routers?
Kent, the "width" of the Internet is a little wider than the NSFnet days. :-) - paul
At 2:57 PM -0800 3/20/98, Kent W. England wrote:
As I recall, the world didn't end when the NSFNET NSSs were installed and they decremented TTL twice. Weren't they changed to decrement once, appearing as one instead of two routers?
You got this part right. I thought we had heard the last of the TTL whining then. Apparently we haven't made much progress in the last eight years.
(For those who don't remember the NSFNET "routers" they consisted of an IBM RS6000 attached to each of several T1/T3 interfaces, interconnected across a FDDI with a route server. Nine RS6000s in all. 200 amp service required. :-)
But this part is wrong. The description above is kind of a combination of the T1 NSS (rack of RTs, each RT services at most a single T1 or Ethernet, they talk to a route processor) and the T3 NSS (RS/6000 with autonomous port cards, pretty similar to any modern router). There's a half-decent description of the NSS architecture at http://www.merit.edu/nsfnet/ Historically yours, --John -- John Scudder email: jgs@ieng.com Internet Engineering Group, LLC phone: (734) 213-4939 x14 122 S. Main, Suite 280 fax: (734) 669-8661 Ann Arbor, MI 48104 www: http://www.ieng.com
On Fri, Mar 20, 1998 at 02:57:04PM -0800, Kent W. England wrote:
(For those who don't remember the NSFNET "routers" they consisted of an IBM RS6000 attached to each of several T1/T3 interfaces, interconnected across a FDDI with a route server. Nine RS6000s in all. 200 amp service required. :-)
You can borrow a frame of our SP/2 to recreate it if you'd like :). Aside from the TTL issue and worrying about interlopers snooping a layer 2.5 topology. Was there any thought put into a tool such as traceroute to follow an Tag/Mpls path. -pee
Kent W. England wrote:
As I recall, the world didn't end when the NSFNET NSSs were installed and they decremented TTL twice.
The world didn't end but I have a vague, cobwebby recollection of having to patch our BSD kernels to use a greater TTL. One of the nice things going on in the IETF right now is that people are rethinking many of their assumptions about what goes without saying. For example, in order for TTL to be effective, not every node needs to decrement it -- we just need to be sure that no loops can form where it doesn't get decremented. Similarly, traceroute was a hack from day one (although a clever one, and over time it even became elegant). If you want to know underlying hops for a particular subnet, use a tool for that subnet type -- you don't need to make everything respond to traceroute. See you ... Scott
participants (5)
-
John G. Scudder
-
Kent W. England
-
Paul E. Erkkila
-
Paul Ferguson
-
Scott Brim