pete@kruckenberg.com (Pete Kruckenberg) writes:
... Customers pay for the network to work end-to-end. More choices mean better performance, more reliability. ...
Not necessarily. Sure, the potential is there. But more than half the time in my experience of multihoming, the pessimal path is chosen. This includes a lot of anycast DNS experiences where each DNS server gives its own close-by mirror server's address as an answer to www.$FOO.com, thus using UDP performance into a predictor of TCP performance. Not only ain't it nec'ily so, it is nec'ily not so. The trivial case of multihoming is inside a campus where a file server, say, might be connected to more than one LAN. Unless you're very careful, your clients will end up talking to the file server through a gateway, that is, to some connection that it has to some LAN other than the client's LAN. This pessimality scales amazingly well to the larger Internet. Don't leave Murphy's Law out of your reliability calculations. More choices can simply mean more points of failure or more opportunities to make bad choices. -- Paul Vixie <vixie@eng.paix.net> President, PAIX.Net Inc. (MFNX.O)
Paul, --On Sunday, 02 September, 2001 8:17 AM -0700 Paul Vixie <vixie@vix.com> wrote:
using UDP performance into a predictor of TCP performance. Not only ain't it nec'ily so [1], it is nec'ily not so [2].
Understand [1] well, but not [2]. Care to elaborate? I'd have thought there were trivial counterexamples (i.e. when it 'is so'), f'rinstance when one path has both huge delay and large packet loss. -- Alex Bligh Personal Capacity
participants (2)
-
Alex Bligh
-
Paul Vixie