Re: draft-heath-ppp-v44-02.txt

> From: jheath@xxxxxxx

> ...
> I think  you are confusing V.44 compression with V.42bis compression. V.44 
> compression (i.e. LZJH) is sufficiently different from LZW that no Unisys 
> or other LZW variant license fees are required for it's use.

oh.

> In answer to your question, V.44 in this environment gets anywhere from 
> 20% to 200% better compression than LZW, and thus BSD PPP, depending on 
> the data.  For HTML, etc. type web pages it is closer to the 200%.  V.44 
> also gets better compression than LZS by ~10%.   With faster execution 
> times.

On which corpus was this measured and with which PPP implementations?
When comparing to LZW, what size dictionary was used, the typical,
barely useful 9 or 10-bit of v.42bis or something reasonable?

I ask because the archives of this mailing list have plenty of marketing
statements of compression performance (made up facts), although not as
many as in the open literature (trade rags).

I also ask because your numbers seem backwards.  LZS as it is in PPP is
a terrible compression scheme compared to others including LZW.  I cannot
imagine something bettering LZS by only 10% but LZW with a reasonable
dictionary by 200%.  For that matter, I cannot imagine anything beating
LZW with a reasonable dictionary size by 200%.  There simply is not that
much entropy in LZW output, unless you do something dishonest such as
using 9-bit codes or clearing the dictionary on every packet.


> The substantial increase in compression ratios that LZJH gets over LZW is 
> the reason the ITU adopted V.44 in the first place to replace V.42bis in 
> modems.
> ...

I don't want to offend anyone, but the ITU is at least as bad as any
other standards organization and frequently (or almost always) makes
decisions for one set of reasons (e.g. patent fees) while claiming to
have other reasons.  In a case like this, what the ITU thinks is
completely irrelevant.  What does matter is measured performance and
the recipes for evalutating and repeating those measurements.

The bigger question remains.  Is the de facto choice of LZS good enough,
bad as it is on compression effectiveness grounds?
Is v.44 enough better to justify adding to the CCP babel?  How many
implementations of v.44 PPP compression can reasonably be expected given
the market dominence of LZS?

Finally, I have a question about the expected status of a v.44 RFC.
It would be Informational, right?


Vernon Schryver    vjs@xxxxxxxxxxxx