Date: Sun, 17 Aug 1997 01:07:45 +0800 (WST) From: Adrian Chadd <adrian@obiwan.psinet.net.au> To: Matt Baker <matt@junior.portal.net.au> Cc: freebsd-isp@FreeBSD.ORG Subject: Re: Multi-homed - Load Balancing - No Single Point of Failure Message-ID: <Pine.BSF.3.95q.970817005910.14827A-100000@obiwan.psinet.net.au> In-Reply-To: <199708161251.WAA12886@junior.portal.net.au>
next in thread | previous in thread | raw e-mail | index | archive | help
On Sat, 16 Aug 1997, Matt Baker wrote: > >From what I've heard, the 2501 won't handle a full BGP load due to the lack > of memory. A couple of questions: It can't handle a *full* BGP load, certainly. 16mb isn't enough for a full BGP feed. > 1. Is it possible to only exist with a limited set of BGP data in Australia? > Is this the iBGP you mention? Yep. I'm connected to telstra internet, and you can ask them for an AS1221-only feed, and they will give you a feed with about 5300 or so routes describing everything in Australia and some japanese stuff too. > 2. Would it be possible to let the 2501 handle the serial traffic on the > link, with a default route to a FreeBSD box which then looks after the > full BGP dataset? The second link could either also hang off the 2501, > or the FreeBSD box. What is your second link coming in via? You don't *need* a full BGP feed unless you want to start doing policy-based routing and some other cool things which most Australian ISPs don't need to ever worry about :) > I really don't have any problem with using FreeBSD boxes as main routers, > but we've got the 2501, so I might as well use it if possible. Believe it or not.. adrian@cortex:~$ uptime 1:46AM up 96 days, 14:21, 9 users, load averages: 0.15, 0.03, 0.01 Thats our FreeBSD box acting as our main gateway. neuro uptime is 1 week, 3 days, 15 hours, 20 minutes System restarted by power-on Thats our 2501 uptime. The only reason its at one week is because it locked up for 5 minutes while I was trying to fiddle redistributing 6000 BGP-learnt routes into OSPF..and then turning them off. NOT a good idea. The only reason I want to use a freebsd box instead of a cisco at this point is because I can build a PC that will replace the Cisco, but have more grunt. The CPU averages around 20-25% during peak, but while doing a BGP update of 5000 routes completely craps itself for a minute or so. Cortex on the other hand, does the update in a second, its a DX2-66 with 16mb RAM. And in case you're worried about memory.. neuro#show ip route summary Route Source Networks Subnets Overhead Memory (bytes) connected 0 4 192 712 static 7 4 852 2912 ospf 1 2 13 720 3684 Intra-area: 0 Inter-area: 0 External-1: 4 External-2: 11 bgp 7585 5283 606 282672 1061780 External: 5645 Internal: 244 Local: 0 internal 183 23790 Total 5475 627 284436 1092878 Thats with 5200ish telstra routes, and 300 local peering routes. People who say unix boxes don't make good routers should be shot. For low to mid end applications, using a PC is cheaper, can be thrown together and got working quicker (how long does it take YOU to get a Cisco 2501 in eh? :), and once running is as stable as hell. In fact, cortex has been just as stable as our Cisco, because all it does is run gated and route. -- Adrian Chadd | "Unix doesn't stop you from doing <adrian@psinet.net.au> | stupid things because that would | stop you from doing clever things"
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?Pine.BSF.3.95q.970817005910.14827A-100000>