
At 07:10 AM 6/9/97 -0700, Bill Manning wrote:
At 09:47 PM 6/8/97 PDT, Randy Bush wrote:
Folk are trying to keep core routers from falling over.
To repeat an earlier unanswered question, what and whose legacy hardware and software is causing the problem?
proteon, ibm, cisco, bay, 3com to name a few, and, to my limited understanding, every implementation of an IP stack. Of course if you are willing to slow down everything and commit to a global network that runs at no greater than 64Kb. (and get everyone else to do the same) then you might be able to get by with the next generation of hardware.
--bill
One of the newbies (Livingston) has BGP in alpha using < 8MB for a full view and a 486 processor without stress. While this won't make a core router, it seems to offer something to consider, even learn from. One of my projects 25 years ago was to reduce the time to calculate the square root on a Seymour Cray machine by a factor of 20 to avoid spending $8 million on a second machine. The revised square root code performed in about the same time scale as the machine's hardware divide function. The clue came from one of the national laboratories and was published in Nuclear Science Abstracts. It took about 6 weeks to do the code and 18 months for the solution to be politically accepted. The original vendor does not always produce sacrosanct stuff. We just went through the process of acquiring our first significant router; one of our main concerns was a router which would allow school districts, libraries and hospitals to benefit from Texas HB2128, which offers distance insensitive T1s and DS3s and was co-authored by our telecomm lobbyist, W. Scott McCollough in Austin. I was somewhat dismayed at the memory limitations of current stuff compared to what I was hearing about the memory requirements. Jonah has more memory on his texas.net usenet news server than you can put on many core routers. Is there anyone working on alternative implementations of software which could possibly solve some of the problems using extant hardware? Will IP multicast help with the usenet stuff? Do we want to continue seeing cams at universities display the local tower while the institution doesn't meet RFC2050 guidelines with respect to utilization of their Class B while the rural areas can't get /19s to support diversity and redundancy? Does Congress need to pass "must carry" legislation similar to the "any willing provider" medical legislation? IMHO, it would be better that some old dogmas and implementations die and be replaced with efficient, robust code and a rather less limiting view of the future. Larry Vaden, founder and CEO help-desk 903-813-4500 Internet Texoma, Inc. <http://www.texoma.net> direct 903-870-0365 bringing the real Internet to rural Texomaland fax 903-868-8551 Member ISP/C, TISPA and USIPA pager 903-867-6571