From a policy standpoint, I would entirely agree. A Tier3/2 cannot afford to be inflexible as can the big boys. Nor do they want to. People many times will pick a Tier 2 over a Tier 1, not necessarily due to
price, but to the flexibility it generally provides. But from a flat-out network design standpoint, I would have to argue that. Even if you only have 3 POP's, if you implement confederations or route reflectors (Just easy examples.) early as part of your network design, you save yourself a hell of a lot of work. If you interconnect your network, and keep your AS exactly what it's supposed to be (I.E. arrogance thread.), and *autonomous* system, then you save yourself work down the road. I'm sure every smaller ISP has clients that stay mainly because they know one or two engineers who will jump through hoops for them if they have problems, and that's just not the case for the big boys. But best practices in network policy and best practices in network design are two different things. Derek -----Original Message----- From: Andy Dills [mailto:andy@xecu.net] Sent: Thursday, August 08, 2002 1:18 PM To: Derek Samford Cc: nanog@merit.edu Subject: RE: NANOG, its decline in s/n On Thu, 8 Aug 2002, Derek Samford wrote:
Personally, every time I post, it's from a Tier-2 perspective. This, honestly, changes absolutely nothing about how I build my network from
logical perspective. There are some minor differences, I.E. I don't own my own fiber, and I don't have many peering relationships. I use transit/transport the same as many other Tier-2's. But the best practices of a Tier-1 are the best practices of any other ISP regardless. Reinventing the wheel is, IMHO, a very bad thing. Over 90% of networking mistakes have already been made, and really, that's what NANOG is for. How many of you out there wish you had done some things different when you look back after rolling out a network? I think
a people
should keep in mind that one of the hardest parts of network design isn't making it work, but making it scale properly. And generally, that's the advice the newer people tend to ignore. Sure other ways will *work*, but they generally won't scale. And the whole point of an ISP is to grow, right?
The whole point of an ISP is to make money. Let's not forget that. Growing has ruined many a fine network. The best practices of a Tier-1 (such a useless term) are NOT neccessarily the best practices for all networks. For instance, a few years ago, I had to bitch at UUnet for three weeks to get them to configure per-packet CEF facing me (3 DS1s). Their first reaction was "No, we don't run proprietary protocols on our network." When I pointed out that I knew for a fact that they were already using CEF switching, cisco-proprietary or not, they finally agreed to try it out as a special circumstance, if it breaks, tough shit. Worked flawlessly for us till we migrated to the DS3 level. Now, it would seem like a reasonable thing for the UUnet's of the world to have such policy, to not run proprietary protocols on their network. (That's why they always turn up circuits with encap frame instead of HDLC.) When you have a network of that size, such sweeping policies are neccessary to maintain sanity. Not so for small networks. It wouldn't make sense for a small network to give up the very flexibility that differentiates it from the large networks. Andy xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx Andy Dills 301-682-9972 Xecunet, LLC www.xecu.net xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx Dialup * Webhosting * E-Commerce * High-Speed Access