[theora] Indexless interpolation search seeking code

[xiphmont at xiph.org]

On Sun, Apr 25, 2010 at 9:10 PM, Chris Pearce <chris at pearce.org.nz> wrote:
> http://www.xiph.org/ogg/doc/oggstream.html says:
>
> "An index is only marginally useful in Ogg for the complexity added; it
> adds no new functionality and seldom improves performance noticeably.
> Empirical testing shows that indexless interpolation search does not
> require many more seeks in practice than using an index would."
>
> I am keen to see seeking code which is almost as fast as seeking with an
> index. Is the code used in this empirical testing online somewhere or
> otherwise available?

That is written based on the assumption of seek operations not being
particularly noticable when physical seek times in the 10-50ms range,
as well as only using exact positioning when it's actually useful.
Even for optical media, something the ffmpeg people like to bring up
constantly, this holds true (after all, MPEG-TS works this way on
DVDs).

In the era of HTTP range-based seeking, I acknowledge that an index is
useful for exact-position seeks, where removing every nonlinear access
is noticable over high-latency links like WiMAX.  For scrubs it still
doesn't matter (those are still just seek to new spot and go).

Monty