Re: Features

On Mon, 24 Nov 1997, Alex Rousskov wrote:

> Hi there,
>
> Just a few numbers for the "compression" discussion. When
> preparing for NLANR Caching Workshop, I did some measurements using logs
> from primary servers and NLANR root proxies.

Good.. I would then guess that the compression debate has occured before.

> As far as I remember, if you compress _every_ file with "gzip -9",
> you get about 20% savings in bandwidth. "Gzip -6" gives about 18%. For
> each file, the minimum of compressed and uncompressed sizes is used. In
> reality, since about 20% of traffic through proxies is not cachable, one
> would get something around 10-15%. There was a research study at DEC that
> reported similar numbers, I think (they also analyzed "delta-compression").

Sounds ok..

> BTW, I would be very careful with estimating the average
> compression ratio based on one site (e.g., www.yahoo.com). A particular
> site may have very good/bad ratios because of its content properties. A
> better test would be to "gzip -c" your cache swap content at night. Still
> not perfect because we are interested in what is transferred, not in what
> is cached.

Very true, I would guess that yahoo is probably showing better compression
then most.. But It was the only big chunk of 'the web' I had localy. I
was mostly concerned with compression/decompression speed..
Received on Mon Nov 24 1997 - 10:55:05 MST