[mythtv-users] Schedules Direct Download Failures

[Mike Hodson]

On Feb 13, 2015 12:33 PM, "Mike Hodson" <mystica at gmail.com> wrote:
>
>I'm personally convinced that a boatload of clients all at the precise
second their NTP synced clocks tell it, are making your 2 DNS servers cry
for mercy.

Actually, its probably not your servers at all...I've thought about this
more and remember now that I've blocked out the horror that is the
standard-rate [read utterly horrid] cable monopoly ISP in my area, whom
I've been thankful to stay away from in my current apartment, if only for
it not actually ever being wired for cable when it was remodeled out of a
1906 department store here in Denver, the old Denver Dry Good company
building in downtown.

I found Aerux, a local WISP with an omni so close even my phone can see
inside my livingroom 2 blocks away, that has wonderful service and direct
unfiltered lines to Cogent and XO. I've forgotten that sometimes,
monopolistic providers will -transparently proxy- all DNS requests through
them, entirely negating the effort on your part to change to opendns or
such.

Even without this mangling, simply by using provider dhcp-assigned name
servers, and not using a third party large cacher like google, opendns, or
level3's resolvers, will cause  the already overloaded ISP servers to sit
as they try to serve your request ... And the schedule run times out
waiting for a slow result.

Hopefully the bad name servers at ISPs will be able to respond seconds
faster to a result they have already from the last time...

Highly suggested: 48 hour or even 72 hour TTLs. Maybe 8 days for those who
infrequently update, once weekly.  Give that overloaded thing a food long
record it can serve back quickly ;). Even hitting disk is faster than
querying a whole recursive root to authoritative lookup.

Also, make sure to set the zone default TTL and not just individual A
record TTLs; I've seen windows-server DNS servers not be quite so intuitive
of allowing just the default TTL to take precedence like is so easy to do
with the right syntax in a bind9 zone file.

Anything you can do to reduce the chance any part of your entire DNS chain,
from authoritative NS server names to the www A record itself, has to be
resolved at all (ie, coming from the cache from yesterday or last week)
during the startup process of a highly parallel, distributed scheduled
process, will help things greatly, and I dare say rather quickly too.
Faster than everyone else updating their schedules and DNS and who knows
what else to lower the chance of a pileup time and slow DNS servers at ISPs.

It seems rfc1912 even advocates for WEEKS of caching for mail records and
the like.

I wonder how much of an effect it will have? I'm curious to know!
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.mythtv.org/pipermail/mythtv-users/attachments/20150213/9fbc3098/attachment.html>