Micro-allocation needed?

From: Joe Abley <jabley@hopcount.ca>
Date: Mon, 21 Jun 2010 17:55:40 -0400

I'm interested in the idea of anycasting one of the pool.ntp.org
herd-members. Every time I've suggested such a thing I've been told
(paraphrasing) that a good (server, client) NTP session exhibits
reasonable RTT stability, this constitutes, in effect, a long-lived
transaction, and hence anycast is not a good answer unless you have
confidence that the potential for oscillations is low, or that the
frequency of the oscillations is very low (i.e. in a private network
this might be a good answer, but across the public Internet it's a
poor answer).

Has the thinking changed, or did I just misunderstand?


This would be better asked on the NTP list, but I'd say it depends on
the accuracy you want to achieve. For the NTP pool, the idea is to try
for good accuracy and very good long-term stability are the goals. That
does not work well of the actual source of the data changes very often.

Aside from losing the advantages of long-term PLL filtering of the time,
you also will see substantial changes in delay (i.e. RTT) and, almost
certainly, jitter.

Unless you are confident that the source of the anycast at any point in
the network will remain stable over a very long term, it really does not
sound like a good solution to me. Then again, with GPS time source
available for <75 USD, anyone who is really trying for really good time
should just buy one and run a local stratum 1 server.

Are you considering doing SNTP or regular NTP?

If regular NTP... I once read some excellent advice on AnyCast:
"It often doesn't make sense to go through the extra complexity in deploying a service with AnyCast addressing if it doesn't justify the benefit."

In this sense, I really don't understand what you will gain.