From: Leif Sawyer <lsawyer_at_gci.com>
Date: Wed, 19 Mar 2003 15:16:02 -0900

Uwe continues the thread with:
> With a 2 MB AJB, you have 1,7 MB free memory for MP3 buffering. The
> 8MB-modified Jukebox has 7,7 MB for buffering and the gain in battery
> runtime is 15-30%.
>
> So if you add dynamic memory allocation so that you can use 1,8 MB
> with a 2MB AJB instead, I think it will give you another percent
> running time or so - alsolutely not worth implementing dynamic memory
> management IMHO.


Well, it's all about cost tradeoff, right?

In this case, the cost is the amount of time to develop the dynamic
memory management scheme. The benefit is scalability for better
buffering (whether on 2Mb or 8Mb) and possibly automatic support for
units with more than 2Mb of onboard RAM.

In order to keep the cost down, we could only implement initial dynamic
memory allocation. This meaning that when rockbox boots up, it figures
out how much ram is available and bases the buffer sizes on the current
ratio.

This benifits both sides, with minimal effort and impact. We're still
not dynamically "tuning" the memory based on the current applications,
so complexity is significantly reduced. This also has the added benefit
of being "ready" for the next HW rev of the AJR/FM/xxx with more than 2M
of ram. (Ok, we don't know if they'll do this, but we'd be ready!)


If this were to get developed by a non-primary developer, would it be
a possibility for mainline inclusion?

Leif

Received on 2003-03-20