[asterisk-biz] Recomendation for DC in NYC for VOIP

[alex at pilosoft.com]

On Wed, 30 May 2007, Mike Hammett wrote:

> Well, that's just it, it doesn't just take the same routes and MPLS has
> nothing to do with it.  What I gather, is that instead of BGP deciding
> the packet path based purely on AS hop count, it tests the destinations
> for latency, packet loss, jitter, etc. and then selects the best
> provider for that given destination.  When provider X has a congested
> peering point (or an administrative peering dispute) you don't have to
> tweak your routers to avoid it, you just get the clean, filtered route
> table (if a BGP customer). While more than provider X, it is
> cheaper\easier than if you were to get provider A, B, C, D, E, F, G,
> etc. and manually tweak it.  While most services don't really care about
> those metrics, they are critical for VoIP, gaming, etc.
If only it was the case. This is *hard* problem. You've described the 
100000 feet overview of how it *should* work. It doesn't quite work that 
way.

I'll give you a simple example: A route announced as /20 that is split 
between multiple locations on provider's network. If you try to optimize 
traffic based on /20 route without knowledge that it is in fact going to 
multiple destinations inside provider's network, you will make performance 
*worse* rather than better. 

Go back to school.

> They are also a great value for low commit users.
Better negotiating skills.

-alex