At 11:00 AM 8/15/95, Dave Siegel wrote:
Let's excuse the fact that gated consumes more memory than a cisco for the same amount of routes for a second...
Okay, so let's talk functionality.
Are there HSSI PCI cards available? Can you do SMDS over this card? Frame? What about a DS3 ATM card, or higher?
And Paul Traina wrote something vaguely similar. I think you guys are both missing what I think was Jon's original point: Routers forward packets faster than PCs, but the forwarding function and the routing protocol function do not have to reside on the same box. You can add a PC (workstation, whatever) which runs the routing protocol and stuffs routes into the router. It doesn't have to support the link-layer du jour. Ethernet will do the job just fine. As I recall the original discussion was of colocating a router, to forward packets, with a workstation, to compute routes. --John
And Paul Traina wrote something vaguely similar.
I think you guys are both missing what I think was Jon's original point: Routers forward packets faster than PCs, but the forwarding function and the routing protocol function do not have to reside on the same box. You can add a PC (workstation, whatever) which runs the routing protocol and stuffs routes into the router. It doesn't have to support the link-layer du jour. Ethernet will do the job just fine.
As I recall the original discussion was of colocating a router, to forward packets, with a workstation, to compute routes.
So what would the normal implementation of such a design be? ebgp-multihop all of your peers into the PC, and then a single peering session the Cisco, presuming no "next-hop-self" routes? I can see some amount of value in such a design, if it could be made to work correctly. Does anybody have the spare equipment to build a lab? (pfeh, yeah, right) Dave -- Dave Siegel Director of Engineering, Net99 http://www.webcity.com/ (602)249-1083 24x7 NOC line http://www.rtd.com/~dsiegel/ (520)318-0696 My Tucson Office
Dave Siegel previously wrote:
As I recall the original discussion was of colocating a router, to forward packets, with a workstation, to compute routes.
So what would the normal implementation of such a design be? ebgp-multihop all of your peers into the PC, and then a single peering session the Cisco, presuming no "next-hop-self" routes?
No. Colocated BGP4 "proxies" (I'm still not sure what to call these, anyone?) would peer via EBGP with other ASes BGP4 "proxies" on the same net. The next_hop BGP4 attribute on all routes exchanged would be that of the routers on the high speed interconnect, not of the "proxies." ASes that do not implement this would still peer via EBGP router-to-router as usual and would not see the "proxies;" eventually everyone would move towards having "proxies" or else router vendors would beef up their products, either way, we're all happy.
I can see some amount of value in such a design, if it could be made to work correctly. Does anybody have the spare equipment to build a lab? (pfeh, yeah, right)
It would work (see the comment on the next_hop BGP4 attribute above). I'm sure there are folk out there that would be willing to experiment, all you need is for two ASes to try it at some exchange (and permission by the exchange to run the necessary ethernet).
Dave
-- Dave Siegel Director of Engineering, Net99 http://www.webcity.com/ (602)249-1083 24x7 NOC line http://www.rtd.com/~dsiegel/ (520)318-0696 My Tucson Office
Nick
So what would the normal implementation of such a design be? ebgp-multihop all of your peers into the PC, and then a single peering session the Cisco, presuming no "next-hop-self" routes?
No. Colocated BGP4 "proxies" (I'm still not sure what to call these, anyone?) would peer via EBGP with other ASes BGP4 "proxies" on the same net. The next_hop BGP4 attribute on all routes exchanged would be that of the routers on the high speed interconnect, not of the "proxies." ASes that do not implement this would still peer via EBGP router-to-router as usual and would not see the "proxies;" eventually everyone would move towards having "proxies" or else router vendors would beef up their products, either way, we're all happy.
Nick
This is very similar to the existing RA route server design. If you want to play with this, then pick up the RS code and try it out. Its a lot closer than the gated base is for doing this kind of "toying around". --bill
bmanning@ISI.EDU previously wrote:
No. Colocated BGP4 "proxies" (I'm still not sure what to call these, anyone?) would peer via EBGP with other ASes BGP4 "proxies" on the same net. The next_hop BGP4 attribute on all routes exchanged would be that of the routers on the high speed interconnect, not of the "proxies." ASes that do not implement this would still peer via EBGP router-to-router as usual and would not see the "proxies;" eventually everyone would move towards having "proxies" or else router vendors would beef up their products, either way, we're all happy.
Nick
This is very similar to the existing RA route server design. If you want to play with this, then pick up the RS code and try it out. Its a lot closer than the gated base is for doing this kind of "toying around".
Indeed. In fact, it is the same idea as that of an RS to some point: the routers on the high speed interconnect use a single BGP4 neighbor to hear and announce routes at the XP. The only difference is that a single organization would run a given RS, but with "proxies" everyone is in full control of their AS. A single RS is nicer in some ways though: it can make a lot of safety checks against the routing policy database it needs to run, but only if the database reflects reality and changes to it are monitored. It's a question of what architecture will be prefered in the end by NAPs and their members; I'd like to give the RA RS a good chance. I was thinking of using the RS code to experiment with implementing some of the weird features I've mentioned in other posts.
--bill
Nick
Ok, so you could not use a proxy server unless you Ogot some other NSP to set up a server also? Is this correct? I want to set up a MAE-East connection, I have a 4000 and only 16 megs or ram. Would I be able to set up a PC as a route server and not need 64 megs or ram in my router? Tue, 15 Aug 1995, Nicolas Williams wrote:
Dave Siegel previously wrote:
As I recall the original discussion was of colocating a router, to forward packets, with a workstation, to compute routes.
So what would the normal implementation of such a design be? ebgp-multihop all of your peers into the PC, and then a single peering session the Cisco, presuming no "next-hop-self" routes?
No. Colocated BGP4 "proxies" (I'm still not sure what to call these, anyone?) would peer via EBGP with other ASes BGP4 "proxies" on the same net. The next_hop BGP4 attribute on all routes exchanged would be that of the routers on the high speed interconnect, not of the "proxies." ASes that do not implement this would still peer via EBGP router-to-router as usual and would not see the "proxies;" eventually everyone would move towards having "proxies" or else router vendors would beef up their products, either way, we're all happy.
I can see some amount of value in such a design, if it could be made to work correctly. Does anybody have the spare equipment to build a lab? (pfeh, yeah, right)
It would work (see the comment on the next_hop BGP4 attribute above).
I'm sure there are folk out there that would be willing to experiment, all you need is for two ASes to try it at some exchange (and permission by the exchange to run the necessary ethernet).
Dave
-- Dave Siegel Director of Engineering, Net99 http://www.webcity.com/ (602)249-1083 24x7 NOC line http://www.rtd.com/~dsiegel/ (520)318-0696 My Tucson Office
Nick
Nathan Stratton CEO, NetRail, Inc. Your Gateway to the World! --------------------------------------------------------------------------- Phone (703)524-4800 NetRail, Inc. Fax (703)534-5033 2007 N. 15 St. Suite B-5 Email sales@netrail.net Arlington, Va. 22201 WWW http://www.netrail.net/ Access: (703) 524-4802 guest ---------------------------------------------------------------------------
Ok, so you could not use a proxy server unless you Ogot some other NSP to set up a server also? Is this correct? I want to set up a MAE-East connection, I have a 4000 and only 16 megs or ram. Would I be able to set up a PC as a route server and not need 64 megs or ram in my router?
I have seen 2501's handle a full routing table with 16 MB's of RAM, and only a single path, but I have some AGS+'s that have 16 that can't hold the full table (it might just need an IOS upgrade) I'm not too sure about the 4000's abilities with 16Megs, but I have one with 32Megs that has full routes, with 2 paths for each route. As for the other NSP, it wouldn't make difference what they were running. If you were already peering with them, you'd have to have the NSP change the IP address of their peer, as well as specify ebgp-multihop... how many NSP's would divulge in this experiment? I don't know. They'd have to trust you with a pretty big stick. If your Unix box is unstable, you can bet they'll shut off peering without thinking twice. Now, the idea would be to take routes from all the providers at Interconnect A, take routes from your routers (maybe another gated box) at the other Interconnect points B, C, D, & E, calculate the best route, let the box flap when links go down, interconnects drop, whatever, and simply send the best route to your little router. Problem here is that a 4000 can't pump all that much traffic. Using a PC as a router would probably provide better performance than the 4000 (though not more than a 4500, 4700, or 7000 series router) Dave -- Dave Siegel Director of Engineering, Net99 http://www.webcity.com/ (602)249-1083 24x7 NOC line http://www.rtd.com/~dsiegel/ (520)318-0696 My Tucson Office
Nathan Stratton previously wrote:
Ok, so you could not use a proxy server unless you Ogot some other NSP to set up a server also? Is this correct? I want to set up a MAE-East connection, I have a 4000 and only 16 megs or ram. Would I be able to set up a PC as a route server and not need 64 megs or ram in my router?
If I remember correctly, MAE-East member gets two host IP addresses (one for a MAE-East connection and one for a MAE-East+ connection. If MFS will let you use one of those two IP addresses for your PC, and if they let you somehow get that PC on MAE-East, then you're ok and you can use the PC as a proxy. But you'll have to talk to MFS. Anyways, I would recommend you upgrade your 4000 to a 4000-M or (better yet) a 4500-M. If you're judicious about who you peer with and what you accept from them and buy transit from someone for what you can't hear directly from your peers, and if you have no other points-of-exit, then you won't be hearing much more than 1x full Internet routing, which takes up ~4MB on a Cisco (just for the paths and routes, not for the software or anything else); a 32MB box will be ok, for now. In the long run? Make you have enough cash to upgrade when it becomes necessary, and you'll be ok :)
Nathan Stratton CEO, NetRail, Inc. Your Gateway to the World! --------------------------------------------------------------------------- Phone (703)524-4800 NetRail, Inc. Fax (703)534-5033 2007 N. 15 St. Suite B-5 Email sales@netrail.net Arlington, Va. 22201 WWW http://www.netrail.net/ Access: (703) 524-4802 guest ---------------------------------------------------------------------------
Nick
So what would the normal implementation of such a design be? ebgp-multihop all of your peers into the PC, and then a single peering session the Cisco, presuming no "next-hop-self" routes?
I can see some amount of value in such a design, if it could be made to work correctly. Does anybody have the spare equipment to build a lab? (pfeh, yeah, right) It's just as 'routing policy server' works. There is such projects and it seems there is some good things in this idea.
But did anybody see real realisation of this?
On Aug 16, 14:34, alex@kiae.su <alex@kiae.su> wrote:
Subject: Re: CIDR FAQ
So what would the normal implementation of such a design be? ebgp-multihop all of your peers into the PC, and then a single peering session the Cisco, presuming no "next-hop-self" routes?
I can see some amount of value in such a design, if it could be made to work correctly. Does anybody have the spare equipment to build a lab? (pfeh, yeah, right) It's just as 'routing policy server' works. There is such projects and it seems there is some good things in this idea.
But did anybody see real realisation of this?
This is effectively the way we do ISDN PRI backup -- the BGP host isn't (necessarily) in the same router that is actually switching the traffic. So it does work, yes. (This is all with Ciscos, BTW.) Yes, PCs and other workstations can route traffic too. We know that. But as has been noted, this has little or nothing to with with the I-D, and probably nothing to do with the CIDR FAQ either. It would be very beneficial if everybody tried to stay on track. -- ====== ___ === Per G. Bilse, Mgr Network Operations Ctr ===== / / / __ ___ _/_ ==== EUnet Communications Services B.V. ==== /--- / / / / /__/ / ===== Singel 540, 1017 AZ Amsterdam, NL === /___ /__/ / / /__ / ====== tel: +31 20 6233803, fax: +31 20 6224657 === ======= 24hr emergency number: +31 20 421 0865 === Connecting Europe since 1982 === http://www.EU.net; e-mail: bilse@EU.net
Sorry, what's the original question? really, we have a great experience with gated and PC based routers there...
At 11:00 AM 8/15/95, Dave Siegel wrote:
Let's excuse the fact that gated consumes more memory than a cisco for the same amount of routes for a second...
Okay, so let's talk functionality.
Are there HSSI PCI cards available? Can you do SMDS over this card? Frame? What about a DS3 ATM card, or higher?
And Paul Traina wrote something vaguely similar.
I think you guys are both missing what I think was Jon's original point: Routers forward packets faster than PCs, but the forwarding function and the routing protocol function do not have to reside on the same box. You can add a PC (workstation, whatever) which runs the routing protocol and stuffs routes into the router. It doesn't have to support the link-layer du jour. Ethernet will do the job just fine.
As I recall the original discussion was of colocating a router, to forward packets, with a workstation, to compute routes.
--John
participants (7)
-
alex@kiae.su
-
bmanning@ISI.EDU
-
Dave Siegel
-
jgs@aads.net
-
Nathan Stratton
-
Nicolas Williams
-
Per Gregers Bilse