IP: some decent comments on "Internet MELTS DOWN AT END"

[Dave Farber <farber () central cis upenn edu>]

Date: Sat, 14 Sep 1996 00:35:42 -0700 (PDT)
From: Michael Dillon <michael () memra com>
Organization: Memra Software Inc. - Internet consulting


On Sat, 14 Sep 1996, Hubert Savelberg wrote:


> What is your opinion/experience  about the published statement in:
>
> http://www.boardwatch.com/mag/96/sept/bwm17.htm


The guy is clueless [ I assume the author Frank Sowa djf] and any ISP that
looks to him for advice is even more clueless.


A) Customers do not call and complain about sites greater than 30 hops
   away because they don't know this. The actual network diameter has been
   shrinking recently due to more exchange points and NAP's in operation
   so the 30 hop limit is highly unlikely to be reached.


B) Yes your upstream might call you and tell you to renumber at any time.
   But you should know this
   ftp://rs.internic.net/policy/internic/internic-ip-1.txt
   and you should be prepared for this
   http://www.isi.edu/div7/pier
   and your provider will give you the time you need to renumber and
   reconfigure things. This is a fact of life everywhere in the world
   if you are a small ISP on today's Internet.


C) Since you know all about renumbering in advance you should already
   have your customers ready to renumber painlessly or they should already
   be insulated from renumbering by using NAT's and RFC1918 addresses.
   Therefore there should be no fallout.


D) If your network connection is not performing as it should, then you
   should know today! not tomorrow. You should be analyzing performance
   on a continuing basis and upgrading as needed to avoid bottlenecks.
   Loopback testing is just plain silly unless your tests show that 
   your T1 can't handle a 1.54mbps to the next hop router.


E) Sprint is not turning away small ISP traffic. They merely have a policy
   that filters traffic from small unaggregated networks. The word
   "unaggregated" is the key here because if your IP addresses come
   from your upstream provider then your small network *WILL* be
   aggregated and Sprint's filters will not affect you.


F) This use of the words "dampen" and "meltdown" are completely out of
   context and have nothing to do with small ISP's at all. When large
   ISP's who run dynamic BGP routing protocols run into problems and
   oscillate between withdrawing and announcing routes, all the tier
   1 NSP's (not just Sprint) dampen their BGP sessions to prevent the
   routers in the network core from being overloaded with too many adds 
   and deletes in the routing tables. Meltdown is a cute way to say
   overload.


G) This thing about trade laws is silly. Trade laws have no effect
   whatsoever on technology and technical capability. If there was
   a law that an airline could not refuse you a seat on an airline
   if you were there an hour ahead of time, would it make any difference?
   No, because when the plane is full, it is full and laws cannot
   change that.


H) He attempts to make it seem as if the IETF is guaranteeing that
   there will be an Internet meltdown this year. Not true. Bob Metcalfe
   is predicting such a meltdown and he is probably an IETF member
   along with thousands of others who do not believe any meltdown is 
   imminent.


I) The Internet has *ALWAYS* been on the verge of collapse and 
   probably always will be. This is better known as the free market
   as opposed to a monopoly market. The telcos have a monopoly 
   so they can make you pay big bucks for an over-engineered network.
   But in a free market situation, the tier 1 NSP's, the tier 2 RNP's
   (Regional Network Providers) and the ISP's at tier 3 only add
   capacity when customers are ready to order and pay for that
   capacity. This is good because it keeps prices under control and
   relatively flat rate.


J) All his 1, 2, 3, 4 points about Cisco routers are either wrong, minor
   problems, or things that have been fixed.


K) Last year people though collapse was imminent when NSFnet shut down.
   But when it actually happened nobody noticed because everything worked
   fine. Later on, however, problems started to appear with route
   announcements as more ISP's started to use the BGP routing protocol
   but some smart engineer invented route dampening and Cisco promptly
   implemented it and the problem was promptly solved just in time. Kind
   of like things normally work in a free market -- see point I above.


L) Of the three URL's given, two are wrong. One should be
   www.ietf.cnri.reston.va.us and the other should be www.internic.net.
   Actually, www.ietf.org is easier to remember. Besides, I don't see
   what any of those URL's has to do with router shutdowns (rare) and
   Internet slowdowns.


M) Now we have a garbled comment about Cisco routers that imply they are
   broken. In actual fact most tier 1 NSP's use Cisco routers because
   they are the only ones that can handle the complex mesh of the
   global Internet core and the traffic load at the core.


N) Some mythical NANOG study is quoted to make some point or other 
   which makes no sense to me. Obviously some reporter looked in on
   a bunch of network engineers talking about stuff that he doesn't
   understand and jumped to conclusions. I suppose he would be upset
   if somebody showed him that there was grease on the engine under 
   the hood of his car! The real world is *NOT* seamless and sweet.
   There are real people, mechanics, engineers, etc, that make things
   works and deal with the grungy mess that underlies 100% of modern
   technical society. If you can't handle this, don't peek under the hood!


O) I suppose I better deal with the specifics rather than just the
   generalities here. In this quote "A problem that Cisco routers have
   with the Internet Protocol itself is causing router update oscillation,
   link/router failures and congestion." it appears that Cisco routers
   have a fundamental flaw at the IP level. However the mention of
   router updates seems to point at BGP which is a higher level protocol.
   Sounds like yet another rehash of the BGP dampening code. In any case
   the important thing here is that engineers are studying what happens,
   discovering some things that don't work well, pinpointing why they
   don't work well, and *FIXING* them. Give the engineers a round of 
   applause for discussing this in the open at NANOG instead of hiding it
   all behind a veil of secrecy.


P) Two more URL's. One so general I don't see the point. The other one
   is broken. Don't these writers even use the Internet?


Q) The column talks about Sprint's route filters as if they target small
   ISP's when in reality they target small networks who also have the
   mistaken idea that they can bypass the address allocation hierarchy and
   still get working addresses. Then it talks about address crowding which
   has nothing whatsoever to do with Sprint's filters. The filters are
   there as part of the impetus to reduce the size of the global routing
   table so it is not filled with garbage like this:


         208.10.16/24 Fred's ISP  --> send to Big ISP
         208.10.17/24 Widget World --> send to Big ISP
         208.10.18/24 Malls Electric --> send to Big ISP
         208.10.19/24 Billy's BBS --> send to Big ISP


   Instead it should look like this


         208.10.16/22 Some BIG ISP customers --> send to Big ISP


   which takes up less global routing table space and still
   gets the traffic where it is supposed to go.
   
R) The columnist does not explain that Sean Doran's quote applies
   to the larger ISP's who are the ones running BGP and who should
   be controlling their routes so that they do *NOT* flap. It is
   punishing poorly run large ISP's and not small ISP's.


S) It seems that Sprintlink customers are treated more leniently.
   So what, they pay Sprint for the service, don't they? If even
   one other tier 1 NSP implemented the same policy as Sprint then
   Sprint's lawyers (who wrote their customer contracts) would
   have the excuse they need to apply the same policy to Sprintlink 
   customers.


T) The columnist mistakenly refers to Yakhov Rekhter as "Cisco Systems'"
   whereas in fact IETF members *NEVER* represent the sompany they happen
   to work for at the time but only represent themselves.


U) The hierarchical IP numbering scheme being discussed is in fact the
   scheme in place today and it has been so for some time. The IETF and
   IANA merely want to document this scheme and clarify it by publishing
   a Best Common Practices RFC so that it is easier for everybody to
   understand and explain what is going one. If this would cause you 
   hardship, tough bananas! That's life. This is how things are in order
   to make the Internet operate effectively and if you didn't know this
   and make engineering and business plans accordingly then that's
   your problem. But it's never too late to educate yourself and to
   adjust your engineering and your policies to lessen the negative
   impact of hierarchical addressing.


V) All this talk about fees for routes is just that. Talk. There isn't
   even an IETF working group yet for this topic but if you are real
   interested you can join piara () apnic net (get ready to be flamed to a
   crisp if you ask dumb questions on this list) or better yet hunt up the
   PIARA mailing list archives at ftp.apnic.net I believe.


W) This thing about "large providers blame small ISP's" is ludicrous. It's
   true that some few employees of large ISP's say nasty things about
   small ISP's but so what. The small ISP's who are clueless and do stupid
   things deserve to have nasty things said about them. There is such a
   wealth of educational material on the Internet about how to run an
   ISP and how to run a network that there really should not be any
   clueless small ISP's. Unless, of course, they think that $9.95
   per month all you can eat service is the road to riches :-(


X) I'm not going to say much more about all this talk of mythical
   charges. Just be aware that anyone can say what they want but that
   does not mean it will happen. But do pay attention to the cost of 
   renumbering. If you do not plan your network and your business
   from day 1 with renumbering in mind then it will hurt bad when you
   have to do it and it could kill your business. Be prepared.


Y) This idea of ISP co-ops is sort of what a tier 2 Regional Network
   provider does. If you are a customer of Netaxs or TLG or IXA then
   not only are you somewhat insulated from a lot of these problems 
   but you have a certain ammount of access to some very skilled
   people who can help you make sure your networks are properly designed
   and configured.


Z) There is no power in owning IP address blocks because at the
   present time IP addresses are not owned. Right now the power
   is in having a *WORKING* IP address block and that is intimately tied
   in to your choice of upstream provider. And if you change providers
   then you will have to change IP address blocks in order to retain
   that power of having a working address.


*sigh*
That guy gets paid for writing his confusing mish-mash of a column
and I get paid nothing for writing this explanation which I have
gone to some trouble in making as accurate as possible.


Michael Dillon                   -               ISP & Internet Consulting
Memra Software Inc.              -                  Fax: +1-604-546-3049
http://www.memra.com             -               E-mail: michael () memra com