November 6, 2002

TiVo Nation

I just got done with my weekly meeting with Ketan. Tyler Johnson did reply to my email but his only response was to schedule another meeting for November 25th at MCNC. So I don't have anything to do between now and then unless the VQM Software is made available for Win32.

So we started talking about what else I might be able to do between now and then. Ketan's had a noticeable obsession with TiVo and a new cache and distribution scheme he calls TiVo Nation. The basic idea is to use a combination of different caching policies based on local degradation with possibilities of reconstruction or P2P caching. In other words, store more things in your cache at lower quality, with the possibility of grabbing stuff out of your neighbors' caches and maybe restoring things to the original quality by combining what you get from other people with what you have.

Anyway, ICME 2003 has put out a call for papers with a December 15 deadline. The paper is only supposed to be four pages. So Ketan's idea was to do some cache policy versus user utility comparisons, given a particular quality/space tradeoff, in an effort to figure out at what point it makes sense to use different caching policies. This is regardless of the P2P or reconstruction aspects described above. This could be simulated with fairly simple mathematical models in a relatively quick manner, and produce enough content for a four page paper.

There are a few different input parameters to consider when putting this model together. There is a quality versus space curve that represents the compression capabilities of an arbitrary codec. There is a quality versus utility curve that represents how much an object is worth to the user at a given quality. There is a content utility distribution (think bell curve) that represents how much an object is worth to the user based on its content. Two additional parameters are needed which are object arrival rate and object consumption rate, representing the object broadcast rate and user's object deletion rate. Some additional variance is added by having the content and/or quality versus utility curves have some margin of error, and by using non-uniform consumption rates.

Anyway, I like Perl and that's what I figure I'll use to put this together. If that proves too slow, then I'll use embedded Perl to compile a C executable that will still let me parse those input parameters extremely easily from plain text files. At least, that's the plan for now.

Posted by josuah at November 6, 2002 9:17 PM UTC+00:00

Trackback Pings

TrackBack URL for this entry:


Post a comment

July 2013
Sun Mon Tue Wed Thu Fri Sat
  1 2 3 4 5 6
7 8 9 10 11 12 13
14 15 16 17 18 19 20
21 22 23 24 25 26 27
28 29 30 31