[mythtv] More scheduling scheduler

[Max Barry]

Just realized discussion has re-opened here.

David Engel wrote:
> We feel pretty strongly that most users don't really need it, even
> if they think they do. 

I'd love devs who feel this way to join in so we can actually thrash
this out. People often seem to talk at cross-purposes about soft padding
-- each not completely understanding what the other is on about.

For example:

Michael T. Dean wrote:
>  On 11/26/2006 09:18 PM, Ian Latter wrote:
>> Last night was another good example though .. Nine ran 45minutes over
>> with 60minutes ... that then shifted two episodes of CSI out by almost a
>> whole episode ... I record 10minutes before a show and 15minutes after,
>> but not even Einstein could account for time distortions like that (Australia
>> would have to be falling into a black hole in order for 60 minutes to run
>> for 105).
>>
>> I'm assuming something similar happened a couple of weeks ago resulting
>> in half of Scorpion King ... never got to the bottom of that one.
> 
> /me wonders how "soft padding" would fix this.
> 
> /me thinks the only solution for Australian (= no valid schedule) TV is
> 1 capture card per channel, a lot of hard drive space, and manual jump/seek 

You are exactly right. However, I'd like to kill this idea that the only
acceptable solution is one that works 100%. This is the same line of
thought that says: "What you really need is a hard buffer." And it's wrong.

If you compared Australian commercial TV guide data to when programs
were actually broadcast, you'd find a probability distribution: most
pretty close to the advertised time, and decreasing numbers of programs
varying by an increasing amount from the scheduled time.

If you want to be 100% sure of capturing 100% of every show you want,
you do indeed need the extreme solution: 1 card per channel, constantly
recording everything. But you could probably get around 99.9% by using a
buffer of maybe 10 minutes before and 90 minutes after each show. You
might get 99.0% with a buffer of 2-minutes before and 20-minutes after.
And so on.

If we had unlimited hard disk space and capture cards, there would be no
reason to embrace any solution that delivered less than 100%. But since
that's not the case, the best solution is to find a sensible balance
between conserving system resources and capturing large safety buffers.
That is, it *is* reasonable to want a solution that isn't 100% effective.

The next question: why not use hard buffers for this? Because they don't
respect the probability distribution. MythTV considers each minute of a
show with a hard buffer as important as every other minute -- it will
even throw a conflict and fail to record another show in order to
capture the last 60 seconds of a hard buffer.

Soft buffers, on the other hand, allow MythTV to distinguish between
"high-probability times" -- the actual scheduled timeslot, when the show
is highly likely to be broadcast -- and lower-probability times, when
the show might be still running but is less likely to be. If limited
resources force MythTV to make a decision between two shows, one of
which is in its high-probability scheduled time and the other of which
is into its lower-probability buffer, MythTV can rightly choose to drop
the latter in favor of the former.

Max.