- NANOG - lists.nanog.org

topological "closeness" (RE: Web o' Wonder)
by Mike Trest 13 May '96

13 May '96

Michael Dillon's Web o' Wonder scenario keys on the ability of smart servers to do an application level re-direction of load to a topographically more appropriate server. Some of our customer's sites are doing a fair job of this by weighting DNS responses to geographically dispersed sites which, hopefully, are topographically dispersed. >Seems to me that this scenario would also benefit from a tool and database >that could determine topological "closeness" even if it doesn't need to >generate filter lists. >If this scenario were easier to implement it could reduce the load of the >major exchange points by encouraging traffic to stay closer to the network >periphery. The CACHE research efforts (see: http://www.nlanr.net and others) indicates that even greater reductions would be obtained if NSP/ISPs increase use of caches for www. Individual campus statistics, the NLANR prototypes, and statistics from CACHE sites deployed internationally suggest that orders-of-magnitude reductions in external network access [i.e. our backbone load] could be achieved. I am implementing a CACHE simply to improve on-net customer satisfaction. I expect the operational costs to be much less than the equivilent backbone or transit costs. If it has a side effect of reducting meet point traffic, then the entire net benefits. I wonder, how many nets would implement such as cache if the effect would be to reduce meet point traffic by orders-of-magnitude? ..mike.. Mike Trest, ATMNET Voice: 619 643-1805 5440 Morehouse Drive Fax: 619 643-1901 San Diego, CA 92121 Pager: 619 960-9070

2 1

Re: topological closeness....
by avg＠postman.ncube.com 13 May '96

13 May '96

Should i drop into Noel-speak mode and start explaining difference between O(n) and O(exp(n))? If something works in several dozens or several hundreds places does not mean it'll work in million places. And even if it can work in million places but can't work with two million places it is only half year worth of growth. --vadim

1 0

Re: topological closeness....
by avg＠postman.ncube.com 13 May '96

13 May '96

>If I could _enforce_ caching by traffic interception then >I could construct vaguely logical cache structures as an ISP. One way to do that is to spoof TCP sessions in the routers :) That is a horrible kludge, though... >But relying on users to configure their end systems with the >identity of the closest cache server using some ill defined >"network metric" is always going to be an erratic affair at best. How about allocating fixed IP address for cacheing servers? Routing will take care of locating the "closest" one, and browser vendors would be able to hardwire the default address. Although one can tell that they have little clue and probably won't do anything about it, given such impressive items on their track record as HTTP 1.0 or Netscape HTML extensions. --vadim

1 0

Re: Worldly Thoughts
by Dave Curado 12 May '96

12 May '96

> ] > Or, possible some small providers buy a multi-megabit circuit from a > ] > large provider who gives them transit. The small provider then connects > ] > at a single NAP and picks up bilateral peering sessions with a bunch > ] > of people there. The result is offloading traffic from their > ] > "transit link", which stands a good chance of being priced as a > ] > "burstable" link. (pay for what you use) That gives the small > ] > provider an economic incentive to operate in this manner. > > ] Quite a few CIX members operate this way. The interesting question in my > ] mind is whether the "big guys" (defaultless nets, for the purposes of this > ] discussion) think that this represents unfair competition or not. > > We've a defaultless net, but I'm not sure that I'm considered a > 'Big Guy'. Hell, we only route 1% of the internet, but maybe if I > lost my aggregates I could be bigger ;) > > The hidden metric that davec above doesn't consider is latency. > > If I peer at a NAP, I forgo the latency my upstream 'multi-megabit > circuit' incurs. Hey, depends on your upstream provider and the NAP you're talking about... for some "Large" providers and at least one NAP, the reverse is true. =-) davec

1 0

Re: Worldly Thoughts
by matthew＠scruz.net 12 May '96

12 May '96

Original message <9605110249.AA04557(a)wisdom.home.vix.com> From: Paul A Vixie <paul(a)vix.com> Date: May 10, 19:49 Subject: Re: Worldly Thoughts > > Quite a few CIX members operate this way. The interesting question in my > mind is whether the "big guys" (defaultless nets, for the purposes of this > discussion) think that this represents unfair competition or not. >-- End of excerpt from Paul A Vixie This is a _very_ interesting question in light of what I just posted... the CIX has the (odd, these days) feature that peering is mandatory for all, and so large providers don't get to choose whether or not they want to haul traffic across the country for these people. -matthew kaufman matthew(a)scruz.net ps. We'd already be at CIX if certain providers had more than a T-1's worth of bandwidth available out of there, specifically to take advantage of the "mandatory peering" situation. Heavily-loaded small pipes to interesting places aren't useful, though.

5 6

RE: Worldly Thoughts
by Jim Browning 12 May '96

12 May '96

>From: Erik Sherk[SMTP:sherk@uunet.uu.net] >> From: David Schwartz aka Joel Katz[SMTP:stimpson@stimpson.igc.net] >> Let's look at a hypothetical situation. ISP1 peers at MAE-E and >> buys transit from MCI there. Now they ask Sprint to peer with them. Let's >> look at how they reach sites on the west coast. > >This is a violation of the rules of Mae-East: it is not to be used for >customer connections. ISP1 must have a private connection to MCI. > >Erik Are you saying that MFS will not allow a private connection between two routers located at the MAE facility, even if the connection is _not_ made using the MAE FDDI switch, but is instead a direct (e.g. HSSI or AIP) connection between the two routers?? If so, I haven't seen any documentation to that effect... -- Jim Browning

4 3

Re: Worldly Thoughts
by Dave Curado 12 May '96

12 May '96

> Let me turn the question around a bit. Suppose N small providers peer > at M places and together represent p% of the Internet. If they simply > appear at one NAP and don't contract for transit, they may reach > 100%-(%p/(M-1)) of the Internet. Is that something to encourage? If > they must contract for transit and as a result reach all major > interconnects, they get 100% (possibly minus a small epsilon for other > reasons). They then don't need to be at any of the NAPs. Are some > providers trying to show up at one NAP only with the aim of not > contracting for transit through anyone even though they can't really > reach others like themselves at a different interconnect? Or, possible some small providers buy a multi-megabit circuit from a large provider who gives them transit. The small provider then connects at a single NAP and picks up bilateral peering sessions with a bunch of people there. The result is offloading traffic from their "transit link", which stands a good chance of being priced as a "burstable" link. (pay for what you use) That gives the small provider an economic incentive to operate in this manner. No comment on whether this is a good idea or bad, but I understand the thinking. davec

3 2

Re: Worldly Thoughts - Cix/SL single T1
by Sean Doran 11 May '96

11 May '96

| The fact that the line hasn't been upgraded could imply that the CIX | isn't of use to SL. It could also say something about SMDS. Sean.

1 0

Re: Worldly Thoughts
by John Curran 11 May '96

11 May '96

At 11:47 PM 5/10/96, Matthew Kaufman wrote: >I understand that nationwide providers don't want to provide me with >"free transit", so, I suggest the following solutions (directed at >those nationwide providers who don't peer with me, and networks like mine): > > 1. Peer with me anyway, but don't accept any of my routes into your > network... that allows me to do the same thing I would do if I were > nationwide and doing quickest-exit (feed all traffic destined for your > nationwide network into it at the west coast) without giving me the > benefit of the return path... this is no more asymmetric than we've got > now, and at least accelerates the one direction. I've got a transit > provider, so I'm going to hand you the very same packets eventually... > you might as well take them off my hands at the interconnect, rather > than have them presented to you by my transit provider. Interesting... it does avoid the transit issue, so it's worth considering to some extent. I'll note that the current west-coast interconnects are not exactly lacking traffic so there may be cases where it may not be advisable to peer anyway if adds more traffic to the interconnect. > 2. Even better, accept my routes into your network but only use them > within region... I know this isn't nearly as easy as #1, so I don't > expect it to happen, but it would be a significant performance benefit > for the local traffic your customers in the region are generating which > is headed towards me. Once outside the region, of course, you ignore the > direct routes from me and just give them to my transit provider. It becomes very entertaining to scale the above model (although it does provide an excuse to use nearly every BGP routing feature in IOS. :-) /John

1 0

Re: Worldly Thoughts
by matthew＠scruz.net 11 May '96

11 May '96

We have a transit connection here _and_ a connection to mae-west. The mae-west connection's best use is to get traffic to and from other medium-sized regional providers without visiting either our or their transit provider (often different ones, so we're saving a lot of hops and potential congestion points here) There's also those nationwide providers (like UUnet) who apparently find it in their best interest to peer with regional networks to take advantage of the increased performance that parties on both sides get when it happens. I think this makes for a lot of happy customers on both sides of the situation, because any increase in performance or reliability between my network and the nationwide network's customers translates into fewer complaints by those customers that "the net's broken today". Sprint had an engineer on the phone with our operations center just a couple days ago, trying to resolve a complaint originating with a Sprint customer that our network was unreachable... had Sprint been peering with us, instead of relying exclusively on our transit provider who happened to be having an outage, Sprint would have had a happier customer and that Sprint engineer wouldn't have had to make that phone call. Last I checked, paying a skilled NOC person for a few hours could pay for months worth of the bandwidth that we'd "get for free" if Sprint peered with us at just a single coast. The unfortunate thing is when regional providers turn into national providers and lose their incentive to peer with other regionals (eg. formerly-BARRNet, who isn't interested in peering with me because I'm only on one coast, even though, with quickest-exit, all the traffic between the west coast part of BBN Planet and our network would be crossing the west-coast interconnect only... seems silly for me to have to go buy a router and a DS3 to the east coast just so regional traffic can stay local and avoid the extra hops imposed by us each using our transit providers) I understand that nationwide providers don't want to provide me with "free transit", so, I suggest the following solutions (directed at those nationwide providers who don't peer with me, and networks like mine): 1. Peer with me anyway, but don't accept any of my routes into your network... that allows me to do the same thing I would do if I were nationwide and doing quickest-exit (feed all traffic destined for your nationwide network into it at the west coast) without giving me the benefit of the return path... this is no more asymmetric than we've got now, and at least accelerates the one direction. I've got a transit provider, so I'm going to hand you the very same packets eventually... you might as well take them off my hands at the interconnect, rather than have them presented to you by my transit provider. 2. Even better, accept my routes into your network but only use them within region... I know this isn't nearly as easy as #1, so I don't expect it to happen, but it would be a significant performance benefit for the local traffic your customers in the region are generating which is headed towards me. Once outside the region, of course, you ignore the direct routes from me and just give them to my transit provider. Meanwhile, the interconnect is still mighty useful for getting traffic around the area, so even without peering with everybody who's there, we see a lot of performance gains that our customers (and the customers of the other networks) appreciate. There's no way we can pick up all of the regional business, and I think most of our competition knows that they're in the same situation, and that's what makes regional peering work. -matthew kaufman matthew(a)scruz.net Original message <199605110144.VAA26028(a)neteng.nis.newscorp.com> From: Dave Curado <dcurado(a)neteng.nis.newscorp.com> Date: May 10, 21:44 Subject: Re: Worldly Thoughts > > > Or, possible some small providers buy a multi-megabit circuit from a > large provider who gives them transit. The small provider then connects > at a single NAP and picks up bilateral peering sessions with a bunch > of people there. The result is offloading traffic from their > "transit link", which stands a good chance of being priced as a > "burstable" link. (pay for what you use) That gives the small > provider an economic incentive to operate in this manner. > > No comment on whether this is a good idea or bad, but I understand > the thinking. > > davec > >-- End of excerpt from Dave Curado

1 0