Subject: Re: [linux-audio-dev] Buffering idea
From: rm (async_AT_cc.gatech.edu)
Date: Mon Jun 17 2002 - 14:27:15 EEST
On Mon, Jun 17, 2002 at 06:29:24AM -0400, Paul Davis wrote:
> >read it at once ?
> >
> >Is this approach safe or useful ?
> >what do you think ?
>
> i proposed the very same idea here last year. its not clear that you
> could totally get around the patent, because at some point, you'd have
> to read some part of the data, and an experienced coder would be able
> to demonstrate that the purpose of the read was to bring data into the cache.
heh. i think we all had this idea. because it's obvious and all.
the nice thing is there's already an interface to do this in the
kernel, madvise(2). so all that needs to happen is that you use one of
the behaviours defined (like MADV_WILLNEED, in conjunction with
mmap).
the actual way to go (i think) if you want to get around this patent
is to setup a kernel level access method similar to the IBM mainframes
where everything is addressed by a 128-bit pointer (including io and
disks). this is just a generalisation of mmap really. then develop a
reasonably smart caching algorithm that takes into account profiles of
past use. you prime the profile at program startup: basically it
will learn that it needs to cache the beginning of the file to satisfy
some realtime heuristic that you set as a policy constraint.
then you just access the samples as data in buffers and don't worry
about it (fuck i'm innovative).
so the only question left is, are you liable if your machine learning
algorithm violates someone else's IP and doesn't tell you? hmmmm.
personally i'm all for willful infringement myself.
(DISCLAIMER: the opinions stated above are not my own, they are those
of your employer, so sue them.)
thanks,
rob
----
Robert Melby
Georgia Institute of Technology, Atlanta Georgia, 30332
uucp: ...!{decvax,hplabs,ncar,purdue,rutgers}!gatech!prism!gt4255a
Internet: async_AT_cc.gatech.edu
This archive was generated by hypermail 2b28 : Mon Jun 17 2002 - 14:34:43 EEST