Re: [RFC] Breaking data compatibility with userspace bzlib

[Jeff Garzik <jgarzik@pobox.com>]

Jörn Engel wrote:
> On Fri, 20 June 2003 12:48:51 -0700, David Lang wrote:
>>he is saying that the memory and CPU requirements to do the bzip
>>uncompression are so much larger then what is nessasary to do the gzip
>>uncompression that the small amount of space saved is almost never worth
>>the cost.

exactly


> just a few).  It is impossible in it's current state to use less than
> roughly 1MB of memory, even though the algorithm doesn't give that
> restriction at all.  Drop that down to 280k, the current zlib value
> and you won't see a difference in compression ratios for jffs2 at
> least.

Well, if you drop that down to 280k, and you do not beat zlib 
compression by a "comfortable margin", it's just pointless code churn in 
addition to swapping out the faster algorithm for the slower algorithm.

If you drop that down to 280k with much better compression than zlib, 
that's fantastic and useful, and people won't mind the slower algorithm :)

The issue of memory usage at high compression rates was the main reason 
why I didn't push the patch upstream and pursue bzlib further.

	Jeff