Tom Schmidt wrote:

I need your opinion on InterNAP. I am currently have a DS3 to my current provider and want to add an additional DS3 for redundancy to the same location. We plan to run BGP4 on both connections.

InterNap has some technology to avoid congested peering points. Does this technology actually work? Isn't it impossible to avoid these peering points? What are your experiences with InterNAP?

If you know what you're getting, then they deliver essentially what you would expect. The problem is that they'd like to lead you to believe that you're getting something magic, thanks to their super-cool ultra-confidential technology that nobody else has. In reality, it's just a computer that scans the BGP tables once a day, throws a few other parameters into the equation, and adjusts prepending and local preferences accordingly. It doesn't have any special provisions to avoid a tempermental route any more than any other BGP speaker. You won't have extra immunity from a fiber cut in Washington because of automated response to fluctuating routes, but you will be at an advantage because you have multiple carriers at your disposal. You, the end-user, have very little control over this. It's the typical BGP problem: you can classify and organize outbound traffic (inbound routes) to your heart's content, but you're very limited in what you can do with inbound traffic (your route announcements). This is probably fine if you're singly homed to InterNAP or if you're dual-homed to InterNAP and a non-Tier 1 NSP. It's probably also fine if you plan on using InterNAP purely as a backup. The argument against InterNAP's philosophy of avoiding public peering is to point out that if you were to multi-home yourself to two of the biggest three or four Tier 1 NSPs that InterNAP buys bandwidth from, you'd probably avoid the bulk of the congested public peering points anyway. (Did you really think that MAE-East was the route of choice between BBN^H^H^HGTEI^H^H^H^HGenuity and AT&T?) In addition, you'd have finer control over your traffic. Based on my observations, between half and two thirds of the traffic you would receive through InterNAP comes through a single NSP that InterNAP's bought bandwidth from, evidence that you might be better off connecting to that NSP directly. (I'm not going to tell you which one it is, but I'm sure you can figure it out if you don't know already.) Outbound, they only prefer that carrier about 40% of the time. (Like I said, you always have finer control over outbound traffic, and InterNAP is no exception.) As far as procedures go, InterNAP is very bureaucratic, bordering on the obsessive. On the plus side, they do seem to be well-organized and responsive, probably owing mostly to the bureacracy. Their engineers seem knowledgable and competent. InterNAP gives you a direct connection to about ten backbones, depending on the POP you connect to. If that's what you were shopping for, then look no further. Mark -- Do not reply directly to this e-mail address -- Mark Mentovai UNIX Engineer Gillette Global Network