on 3/15/09 4:53 PM, Robert Hajime Lanning said:

> So?  And how many people work in that environment?  Yes, it is there,
> and a lot of people use the system.  But, it is invisible to the users.

How many people use the root nameservers?  That's the size of the 
potentially affected community.  Even a tiny fraction of one percent of 
the daily traffic could easily be millions or billions of queries, and 
the potentially affected user community could easily number in the 
thousands, tens of thousands, hundreds of thousands, or millions.

> Usually when you do use anycast, it is on-top of a standard
> NAT/next-hop/proxy based cluster.  (And to be clear, I said "Usually"
> not "Every".)

I'm sure that is your experience in this area.  However, others may well 
have other experience regarding the nature or frequency of the 
implementation of the technique.  When you and they both collapse the 
mental tag down to just "anycast", you're throwing away a great deal of 
information that could clearly demonstrate that your situation 
completely different from theirs.

I can tell you that there are at least three or four different root 
nameserver operators who all say that they use anycast as part of their 
mechanism for distributing the load, but I can also tell you that each 
of those groups use a different implementation method.  When they talk 
to each other, they know enough about the situation to ask the other 
side to remind them of their implementation details (if they've 
forgotten), so that they can more accurately share notes.


This is a technique that is complex enough, and can be implemented in 
enough different ways, that you need to be more specific when you 
discuss the use of this method with respect to certain types of 
protocols -- especially protocols that use UDP because it's much lighter 
in weight, but where a great deal of state information is maintained 
over very long periods of time using complex statistical methods, 
despite the fact that no particular given packet may be delivered reliably.

> No, but if someone wants to do it, I would say "go ahead, but here are
> the caveats..."  It might work enough for there needs.

So what are the caveats?  Do you know all of them?  Have you shared any 
of that information with anyone when discussing this subject?


Joe Abley is a very talented network engineer, and my recollection is 
that he has been one of the most experienced persons at ISC with regards 
to their use of anycast for f.root-servers.net.

However, this expertise does not actually translate to use with NTP, 
regardless of whatever you think he says in his article "Fear and 
Loathing in the Routing System" published in the February 2008 issue of 
_;login:_ magazine, which you can read at 
<http://www.usenix.org/publications/login/2008-02/openpdfs/abley.pdf>.

In fact, it was his article with all of its serious mistakes, that lead 
me to write my article for _;login:_ magazine.  I don't know nearly so 
much about anycast, but I do know more about NTP, and I know enough 
about anycast to have a pretty good idea of some of the serious risks 
with regards to NTP.

> Ok, so what is the purpose of the LOPSA "Tech" mailing list?  Since it
> seems that all Tech topics have their own lists.  <sarcasm>I guess we
> should disband this list.  And, everyone should go a subscribe to the
> 150 lists pertaining to all the topics of interest.</sarcasm>

There are plenty of areas where lots of people have wide and useful 
experience with a given program or protocol, and most any of those areas 
should be perfectly suitable for discussion on this list.

However, NTP is an area where it seems like ~99% of the people who talk 
about it understand precisely zero of how it actually works and how it 
should be architected, and they tend to have a lot of myths that they 
continue to propagate about it to everyone they talk to.

Part of why I'm here is to stop these myths from being re-propagated yet 
once again, and to redirect people to a resource where there is a wide 
collection of knowledgeable individuals who are happy to share their 
experience with others, and where they are relatively likely to share 
the correct information in the correct way.

> These topics are not outside the list's stated topic.  Just because
> there is a "better" place to discuss a specific topic, does not make
> this an inappropriate place to also discuss it.  It is inappropriate
> when it becomes off-topic for the list.

It's not just a matter of whether it's on-topic or not.  It's also a 
matter of whether the information to be given is likely to be correct, 
and whether people casually toss around concepts like NTP via anycast, 
without properly covering the caveats.

Note that this is not the multicast or manycast techniques, this is the 
use of plain anycast.

-- 
Brad Knowles
<[email protected]>        If you like Jazz/R&B guitar, check out
LinkedIn Profile:                 my friend bigsbytracks on YouTube at
<http://tinyurl.com/y8kpxu>    http://preview.tinyurl.com/bigsbytracks
_______________________________________________
Tech mailing list
[email protected]
http://lopsa.org/cgi-bin/mailman/listinfo/tech
This list provided by the League of Professional System Administrators
 http://lopsa.org/

Reply via email to