On Wed, Dec 3, 2008 at 1:18 PM, Scott Brim <[EMAIL PROTECTED]> wrote:
>> It need only expire the cached map entry for inactivity. Which it will
>> tend to do over tens of minutes with no use
>
> Correct me if I'm wrong but it seems to me that whatever higher
> stratum NTP servers are used will be (1) well used, and (2) in popular
> prefixes.  What servers do you use?  My little MBP uses apple.com and
> nist.gov.  I doubt that they would be removed from the cache.


In Red Hat Linux, the default installation is:

server 0.pool.ntp.org
server 1.pool.ntp.org
server 2.pool.ntp.org

pool.ntp.org consists of a large, random assortment of volunteers who
have decided it would be cool to run a public NTP server. Two lookups
over time get two more or less randomly chosen sets of IPs:

nslookup 0.pool.ntp.org
Server:         10.187.99.2
Address:        10.187.99.2#53

Non-authoritative answer:
Name:   0.pool.ntp.org
Address: 208.38.65.35
Name:   0.pool.ntp.org
Address: 4.79.132.221
Name:   0.pool.ntp.org
Address: 64.247.17.255
Name:   0.pool.ntp.org
Address: 65.255.217.202
Name:   0.pool.ntp.org
Address: 72.167.54.201

nslookup 0.pool.ntp.org
Server:         127.0.0.1
Address:        127.0.0.1#53

Non-authoritative answer:
Name:   0.pool.ntp.org
Address: 65.111.164.223
Name:   0.pool.ntp.org
Address: 65.255.217.202
Name:   0.pool.ntp.org
Address: 69.31.43.10
Name:   0.pool.ntp.org
Address: 72.167.54.201
Name:   0.pool.ntp.org
Address: 75.144.70.35


Bear in mind that an individual ntp client in normal operations backs
off to 17 minutes between packets and can be configured to go more
than a -DAY- without sending a packet.


> If a
> site communicates with zillions of others, so there is a lot of churn
> in the cache, it probably wants to reconsider its caching strategy in
> the first place, but second, steps can be taken to make sure a map for
> a particular prefix doesn't get squeezed out.  (That doesn't mean it
> doesn't get refreshed as TTL expiration approaches.)

Here's a free solution:

Don't remove entries from the cache unless you run out of memory.
After the short, perhaps 30-second TTL expires, the next packet you
send to that destination triggers a map refresh in order to statisfy
http://bill.herrin.us/network/statechange.html . The results of the
refresh will, if necessary, delete the old entry. While you wait for
the results of the refresh, keep using the expired map. 99.9% of the
time this will be the correct action. For the rest, its a normal,
everyday black hole anomaly that clears in a few hundred milliseconds
when the refresh comes in.

The hitch is that lisp sits rather closer to the core than most of the
pull-cache solutions that have been discussed. Are you sure you're
going to have enough cache to keep 20 or 30 minutes worth of *all* the
destination addresses that the site's folks have sent packets to? I
think I'd want to see some metrics on how many unique destination
addresses leave the biggest site that would be behind just one ITR.

Regards,
Bill Herrin


-- 
William D. Herrin ................ [EMAIL PROTECTED]  [EMAIL PROTECTED]
3005 Crane Dr. ...................... Web: <http://bill.herrin.us/>
Falls Church, VA 22042-3004
_______________________________________________
rrg mailing list
[email protected]
https://www.irtf.org/mailman/listinfo/rrg

Reply via email to