Classical Cat

Off on another track on this one - found this looking for some music to listen to at work while doing a project. Classic Cat has links to tons of free classical music recordings and is well worth checking out.

My current obsession is Chopin - I re-discovered a Prelude (Op 28 - 20) that I'd stumbled across in college and always loved, but lost the recording of it and couldn't remember the name. There are several different recordings of it available and it's very interesting to hear the different interpretations.

Anyway, good stuff and worth checking out.

BGP #2

BGP main attributes - Section One

BGP, as mentioned before, was built with several primary attributes in mind. The first of which is scale. BGP is the routing protocol for the whole Internet, essentially, and so it needs to be able to scale like crazy. How much, you ask? Check this out - as of this blog entry, you're looking at around 450,000 routes. That would melt any normal IGP - but it also means that BGP has to give up some things too. Keep in mind that a lot of people will have multiple copies of the BGP table in their router at a time - one for each ISP that they connect to.

As mentioned before, one of the things that BGP has to give up is rapid convergence. In order to scale up, BGP has to place a lighter load on the routers that run it, and it does this by

1. Never sending periodic updates - you get one update when a route changes - you better write it down
• Pretty much every other protocol does a periodic refresh of routes or link state
2. Those updates, since you never get another one, are sent reliably over TCP to guarantee delivery
• Yes, TCP is "heavier" than UDP or direct IP encapsulation, but it's a set up once and forget it sort of thing with BGP - the TCP session stays up as long as BGP does, which could be literally years
3. Updates are triggered, batched, and generally rate-limited
• Since they're not periodic, updates are only sent on a "triggered" event or change
• Updates are saved up and sent in batches to peers
• Cisco defaults to 5 second batches for internal peers and 30 second batches for external peers

Put all of this together and you get a protocol that tries to place a low load on the hardware, is very stable, but gets around to convergence - or figuring out what routes are valid for a given destination - in a more pokey fashion than other protocols. It's all part of BGP's charm.

I'll get back to BGP at some point here, but I thought I would mention this interesting find. While I was skulking about looking for some good CISSP materials, I found this podcast, and found the one episode I heard very interesting as it was discussed from the point of view of experienced TAC engineers. The episode I grabbed was about GETVPN and DMVPN and was quiet good. Definitely worth checking out if you like the intersection of security and Cisco.