John This is a good suggestion. We consciously opted for Arista 7050 switches over full-table routers a good few years ago (2016) and made up the difference with selective route download. EOS has a really cool feature where a prefix list can be set to a HTTP endpoint so we have internal anycast endpoints whereby each switch can dynamically update its prefixlist, and those are managed by sflow analysis. IIRC the busiest edge has about 5k prefixes installed (plus default) and it has worked flawlessly with newly relevant routes being installed in a few minutes. We've moved on and up a few generations hardware wise now but the basic operation is still the same.
cheers Simon photo ( https://simwood.com ) Simon Woodhead CEO, Simwood *Phone* +44 330 122 3021 ( tel:+44+330+122+3021 ) | *Mobile* +44 7976 238 487 ( tel:+44+7976+238+487 ) *Website* simwood.com ( https://simwood.com ) | *Email* simon.woodh...@simwood.com ( simon.woodh...@simwood.com ) linkedin ( https://www.linkedin.com/in/simonwoodhead/ ) twitter ( https://twitter.com/simwoodesms ) IMPORTANT: The contents of this email and any attachments are confidential. They are intended for the named recipient(s) only. If you have received this email by mistake, please notify the sender immediately and do not disclose the contents to anyone or make copies thereof. On Wed, Jun 28, 2023 at 21:46:08, Will Hargrave < w...@harg.net > wrote: > > > > Hi John, > > > > Why not simply accept fewer routes (plus a default) into the existing > Arista EOS BGP and so the hardware FIB? Then you can actually take > advantage of the hardware forwarding. > > > > With this setup you’re using the relatively slow control plane (the Intel > FM6000 was released a decade ago and I can’t imagine Arista paired it with > a super-fast SoC…) to route and that won’t work very quickly at all. In > fact it may not have enough RAM and CPU to effectively deal with a modern > full table, it would be better to just use a modern 1U server for this. > > > > Will > > > > On 28 Jun 2023, at 21:21, John P Bourke wrote: > > > >> >> >> Hi >> >> >> >> >> >> >> >> I may have “an” answer. I think the Americans call this a “Hail Mary >> Pass”. >> >> >> >> >> >> >> >> I have a bunch Arista 7150s, which are EOL and a disappointment. But I >> found this. >> >> >> >> >> >> >> >> https:/ / research. kudelskisecurity. com/ 2015/ 10/ 01/ >> hacking-arista-appliances-for-fun-and-profit/ >> #comments ( >> https://research.kudelskisecurity.com/2015/10/01/hacking-arista-appliances-for-fun-and-profit/#comments >> ) >> >> >> >> >> >> >> >> The Arista runs a full Centos 7.6. You strip out the Arista BGP process >> and BIRD (or FRR I guess) and you have a route server. I say route >> server, because by pulling the Arista BGP process you have no interaction >> with the RIB. >> >> >> >> >> >> >> >> Thanks >> >> >> >> >> >> >> >> John >> >> >> >> >> >> >> >> BTW – Not dissing Arista. The 7150 is a bit of a unicorn in their >> portfolio, using a chipset from Intel which they bought from a startup, >> which Intel then dropped so Arista understandably did not put a lot of >> effort into beyond the High Frequency Trading use cases that this low >> latency switch is good for. >> >> >> >> >> >> >> >> >> >> >> >> *From:* Tim Bray < tim@ kooky. org ( t...@kooky.org ) > >> *Sent:* Wednesday, June 28, 2023 6:56 PM >> *To:* uknof@ lists. uknof. org. uk ( uknof@lists.uknof.org.uk ) >> *Subject:* Re: [uknof] Full table routers >> >> >> >> >> >> >> >> >> On 28/06/2023 10:27, John P Bourke wrote: >> >> >> >>> >>> >>> Any recommendations for full table routers. We don’t need more than 10G. >>> >>> >> >> >> >> I used Debian + FRR on HP proliants. With startech Nics with intel >> chipset. Unusual, but did the trick. Help that there was a whole >> stack of the same hardware running services in the same place. They >> take a while to boot, but you can make it faster and I think the newer >> variants are better. >> >> >> >> >> >> >> >> Software wise, takes a bit of getting used to. Sometimes conflict >> between FRR and what Debian wants to do for network setup. Also you >> can use CAKE :) Also run any scripts or monitoring you want onboard >> (like counting the BFD flaps per hour to watch the problems that go away >> and come back very quickly) >> >> >> >> See also distributions that bundle FRR more specifically for networking >> rather than a general distribution. >> >> -- Tim Bray Huddersfield, GB tim@ kooky. org ( t...@kooky.org ) +44 >> 7966479015 >> > >
