Non-Routing BGP Speakers
RFC 1771 says in part: The hosts executing the Border Gateway Protocol need not be routers. A non-routing host could exchange routing information with routers via EGP or even an interior routing protocol. That non-routing host could then use BGP to exchange routing information with a border router in another Autonomous System. The implications and applications of this architecture are for further study. Would anyone be willing to share their experiences with (or thoughts about) this approach? Given that 64 MB of RAM for routing tables woudl cost only around $2,000, this seems like a totally sensible way to build a small, multi-homed AS. Will finding a vendor-supported system for this be ... difficult? (I'm not exactly sure whether a BSD box running Cornell GateD counts as "vendor-supported". ;-) -- Sean Shapira sds@jazzie.com +1 206 443 2028 <a href="http://www.jazzie.com/sds/">Sean's Home Page</a> Serving the Net since 1990.
In message <m0s4eEt-000OXcC@jazzie.com>, Sean Shapira writes:
RFC 1771 says in part: The hosts executing the Border Gateway Protocol need not be routers. A non-routing host could exchange routing information with routers via EGP or even an interior routing protocol. That non-routing host could then use BGP to exchange routing information with a border router in another Autonomous System. The implications and applications of this architecture are for further study.
Would anyone be willing to share their experiences with (or thoughts about) this approach? Given that 64 MB of RAM for routing tables woudl cost only around $2,000, this seems like a totally sensible way to build a small, multi-homed AS. Will finding a vendor-supported system for this be ... difficult? (I'm not exactly sure whether a BSD box running Cornell GateD counts as "vendor-supported". ;-)
-- Sean Shapira sds@jazzie.com +1 206 443 2028 <a href="http://www.jazzie.com/sds/">Sean's Home Page</a> Serving the Net since 1990.
BSDI works and comes with gated, though not the latest. The Riscom-N2 is supported by BSDI and can give you 2 56k or even T1 lines speaking Cisco HDLC or PPP. I can't say I've ever tried it, but some people say it would all work fine. You could also take a subset of full routing since you probably won't be doing transit between major providers. We have hosts running gated to gather routing statistics. Our routers are also running gated. They are RS6ks running custom interfaces and a modified AIX kernel. 64MB RAM works for us. Many of our machines have 32 MB of RAM. PSC runs DEC Alphas with gated. This is a supported but expensive option. Handy if you have a Cray and don't want a wimpy router in your FDDI path. Avoid sys5 based stuff like Solaris and SGI Irix since it can't do CIDR routes and barfs badly on overlapped routes. Probably HP too. That leave DEC OSF and AIX. You could probably go with NetBSD on an older Sun. For PCs there is BSDI, FreeBSD, Linx. If you can afford to be dual homed, you probably can afford a router rather than a PC serving as a router. I'd love to hear how things go if you go with BSDI. There is also an Emerging Technologies T1/56k card that claims ISDN LAPB, FR, X.25, Cisco HDLC, PPP and which sounds great on paper but I haven't heard any user testimonials yet. Curtis
On Thu, 27 Apr 1995, Curtis Villamizar wrote:
Would anyone be willing to share their experiences with (or thoughts about) this approach? Given that 64 MB of RAM for routing tables woudl cost only around $2,000, this seems like a totally sensible way to build a small, multi-homed AS. Will finding a vendor-supported system for this be ... difficult? (I'm not exactly sure whether a BSD box running Cornell GateD counts as "vendor-supported". ;-)
BSDI works and comes with gated, though not the latest. The Riscom-N2 is supported by BSDI and can give you 2 56k or even T1 lines speaking Cisco HDLC or PPP. I can't say I've ever tried it, but some people say it would all work fine. You could also take a subset of full routing since you probably won't be doing transit between major providers.
Emerging Technologies also makes sync cards with drivers supported under BSDI, FreeBSD and some forms of SysV UNIX. They have been discussed on either (or both) the inet-access and bsdi-users lists in the past. Archives for inet-access are at earth.com (or is that ftp.earth.com) and for bsdi-users at ftp.bsdi.com. There is sometimes a search engine available for bsdi-users from a link at http://www.bsdi.com. I got my info by emailing dennis@et.htp.com but you could phone (516) 271-4525 or fax (516) 271-4814 So there are at least two possibilities for building 80x86 boxes into routers by using off-the-shelf sync cards and UNICES.
with NetBSD on an older Sun. For PCs there is BSDI, FreeBSD, Linux.
I believe that support for sync cards under Linux is fairly new. Tread carefully there.
If you can afford to be dual homed, you probably can afford a router rather than a PC serving as a router.
There is also the question of support, spares, previous knowledgebase etc. Build-your-own isn't for everyone but it is nice to have a choice. Michael Dillon Voice: +1-604-549-1036 Network Operations Fax: +1-604-542-4130 Okanagan Internet Junction Internet: michael@junction.net http://www.junction.net - The Okanagan's 1st full-service Internet provider
On Thu, 27 Apr 1995 22:49:34 -0400 Curtis Villamizar wrote:
If you can afford to be dual homed, you probably can afford a router rather than a PC serving as a router. I'd love to hear how things go if you go with BSDI. There is also an Emerging Technologies T1/56k card that claims ISDN LAPB, FR, X.25, Cisco HDLC, PPP and which sounds great on paper but I haven't heard any user testimonials yet.
The RIPE NCC (AS3333) is using a BSDI box as BGP router, so that definitely works. Make shure you get decent hardware; PC manufacturers sometimes cut corners and, for instance, I had to disable the external cache on amsterdam.ripe.net to make the box stable. We're not using serial cards as we were able to attach to the Amsterdam DMZ directly. (and you can play doom on a PC; try that on a cisco ;-) I have been thinking about Sean's idea of using gated as BGP talker, and haveing only gated talk to the box switching packets. The routes sent to the router can then be aggressively aggregated only based on nexthop, because the router doesn't need more detailed routing information. I haven't further investigated the idea, but I think it might be possible to e.g. run CSC-3's with 'full' routing this way (if in Europe, you prefer one carrier for US routes, then 198/7 might be possible...) Unfortunately, I don't have the time to (test-)implement it. Note that this might scale better than using router-CPU's; it is much more likely that the generic computer industry can scale up CPU's faster than dedicated router manufacturers can. A disadvantage of this approach is that you are running something different than the majority of BGP talkers, and there might be incompatibilities. This quite clearly showed during the Danvers IETF when someone inject ASpaths with 109 109 109 on the end; this caused older gated's to shut down the peering and caused quite some problems for those sites using gated. We have upgraded our box since then; Curtis, have you had a chance to look at this yet? It is a disaster waiting to happen again, and next time the players might not be convienently located in the same IETF terminal room... Geert Jan
I recommend <URL:http://www.vix.com/pc-hw/> to anyone considering the BSD/Gated solution, either for a conventional router (which is what I use for my Alternet/BARRnet connectivity and it works quite well) or for a discrete routing protocol machine where the switching machine is elsewhere (as is being discussed here presently.) PC hardware can be made as reliably as any other kind, but you can't buy the cheap stuff -- my www page above has some recipes that are known to work. Buying something off the shelf somewhere won't work. A 66MHz PC running BSD/Gated fills its routing table (thus completing the initial phase of a 27,000 route BGP4 session) in about 1/3 the time a Cisco 7000 does it. On the flip side, it can only route about 600 packets per second (quite adequate for my T1 line but obviously woeful if put into an Ethernet or T3 path). 600/sec is with "screening" turned on, it's more like 1,000/sec if you don't need any forwarding security, which is to say never. Port density isn't very good, either -- I can only put about 8 T1's into a box before I have to start a new box, which is a lot less than what a Cisco 7000 can do. On the other hand I can buy new boxes for less than the fractional cost of that 7000, and then tie them all together with an Ethernet. That comes down to a matter of taste, and most people (myself included) prefer the no-moving-parts angle of the Cisco, even if its higher horsepower isn't always called for.
PSC runs DEC Alphas with gated. This is a supported but expensive option. Handy if you have a Cray and don't want a wimpy router in your FDDI path.
Expensive in terms of people, perhaps, in that it doesn't run out of the box. But once the config is made, it is fairly easy to maintain and has some nice properties that you can't easily find on vendor routers. They also tend to have bigger queues than vendor routers. On the downside, the newest generation of routers blows the doors off of the workstations, especially in terms of packets per second. (For example, the last years Alphas push something in the 3000-5000 pps range, where the vendor routers go 20-100 kpps. Don't know what this year's model does. Note that 3000 pps is full FDDI if people use reasonable -- 4kB -- packet sizes, though). The real rebuttal I wanted to make, though, is that the Alpha workstation costs in the $10K price range, with about $1200 per interface for FDDI (less if you can do the UTP business which alas we cannot). Compare this with todays prices on vendor routers and then tell me if you think it is expensive...
Avoid sys5 based stuff like Solaris and SGI Irix since it can't do CIDR routes and barfs badly on overlapped routes. Probably HP too. That leave DEC OSF and AIX. You could probably go with NetBSD on an older Sun. For PCs there is BSDI, FreeBSD, Linx.
I should add that we haven't extensively tested the CIDR stuff on our Alphas since we don't currently take full routing tables. But if Curtis says it should work I believe him :-) --Jamshid
In message <9504281836.AA21138@mailer.psc.edu>, "Jamshid Mahdavi" writes:
I should add that we haven't extensively tested the CIDR stuff on our Alphas since we don't currently take full routing tables. But if Curtis says it should work I believe him :-)
Uh.. thruth in advertising time. There were some radix tree bugs in BSD that Dennis found when playing with NetBSD on Sparc and BSDI. I'm not sure if they made it into BSD 4.4Lite, BSDI 2.0, or any DEC OSF product. I think Jeff sent the patches to someone at UCB and to BSDI. I think they made BSDI 2.0 but I'd guess not the others yet. Better get a more reliable source. :-)
--Jamshid
Curtis
Uh.. thruth in advertising time. There were some radix tree bugs in BSD that Dennis found when playing with NetBSD on Sparc and BSDI. I'm not sure if they made it into BSD 4.4Lite, BSDI 2.0, or any DEC OSF product. I think Jeff sent the patches to someone at UCB and to BSDI. I think they made BSDI 2.0 but I'd guess not the others yet.
Better get a more reliable source. :-)
Yikes. That may be me. All of the bad stuff Dennis and Jeff found in BNR2 (which formed the basis for BSD/OS 1.x) was sent to CSRG and most of it appeared in 4.4BSD-Lite (which formed the basis for BSD/OS 2.0). Some additional things have been fixed since then, according to Keith Sklower and others, and we should see some additional improvements in 4.4BSD-Lite-2, due out any day. Elise, perhaps Merit should invite Jeff Honig <jch@gated.cornell.edu> of the GateD Consortium to attend the next meeting and give us an update? I could sort of do it but I'd be hacking around in the dark -- Jeff actually knows what's going on. It's pretty clear that GateD is an important tool for North American (and other) Network Operators.
From: Paul A Vixie <paul@vix.com> Subject: Re: Non-Routing BGP Speakers
Uh.. thruth in advertising time. There were some radix tree bugs in BSD that Dennis found when playing with NetBSD on Sparc and BSDI. I'm not sure if they made it into BSD 4.4Lite, BSDI 2.0, or any DEC OSF product. I think Jeff sent the patches to someone at UCB and to BSDI. I think they made BSDI 2.0 but I'd guess not the others yet.
Better get a more reliable source. :-)
Both FreeBSD and Lite-2 have the bugfixes for mask sorting. I assume NetBSD will carry it the moment that Lite-2 is made public.
participants (7)
-
Curtis Villamizar
-
Geert Jan de Groot
-
Jamshid Mahdavi
-
Michael Dillon
-
Paul A Vixie
-
Paul Traina
-
sds@jazzie.com