[Zlib-devel] Google's version of zlib code

[Arjan van de Ven]

For all the "deflate is dead".... deflate, but more importantly,
inflate are comparatively cheap in terms of CPU usage (and thus
power). Yes XZ gets an incrementally better ratio, but only
incrementally for many cases (esp for files that are not very large)
and ends up quite a bit more expensive on either side.

The google thing shows that there is room on the "get more expensive"
side, almost to the insane level.
But there is also room on the "get cheaper" part, and at work we've
been playing with zlib performance lately (hopefully to go open soon
once we're a bit further along).

Getting zlib's deflate 25% faster hasn't proven to be that hard, and
line of sight for a 2x to 3x speedup is there (e.g. early prototype
code).

We haven't looked at inflate much, beyond making the alder32 code (is
that used much?) 2x faster, but I'm sure there's some room there too.



On Sun, Mar 3, 2013 at 10:13 AM, Greg Roelofs <newt at pobox.com> wrote:
> Mark Adler wrote:
>
>> On Mar 2, 2013, at 7:54 AM, Nelson H. F. Beebe <beebe at math.utah.edu> wrote:
>>>      Google publishes Zopfli as open-source compression algorithm to speed up Web download
>>>      Compression is about 100 times slower than conventional methods but compresses about 5% better,
>
>> That's cool, but it seems like an awful lot of effort for a small improvement.
>
> Agreed.  Other than a tiny, tiny handful of purveyors of static content--
> someone like Akamai, perhaps, except they're probably contractually bound
> not to screw with it and probably don't have the compute resources to waste
> on it anyway--I can't imagine too many entities that would find this useful,
> Google included.
>
>> Perhaps it's time to add a better compression method to HTTP's accept-encoding.
>
> LZMA seems like an obvious candidate.  There's probably room for a QuickLZ or
> Zippy or similar on the other end (faster/worse), too.
>
> Greg
>
> _______________________________________________
> Zlib-devel mailing list
> Zlib-devel at madler.net
> http://mail.madler.net/mailman/listinfo/zlib-devel_madler.net