Re: NetCache FAQ (NOTE: NetCache give away ends September 30)

From: James R Grinter <>
Date: Tue, 23 Sep 1997 15:20:21 +0100

(cc'd to ircache. It might be more appropriate to discuss this there,
though there is no general proxy cache users list that I'm aware of.)

On Tue 23 Sep, 1997, Robert Thomas <> wrote:
>Personally, I think Linux and FreeBSD would be THE most popular OS's for
>caches, from small to allmost-huge.. Didn't I hear someone bandy about the
>number of 80% of all proxy servers on the net are running unix/squid? And,

Actually, this one is really difficult to work out. For a start, Squid
1.1 doesn't easily give away its version in request strings, as it
hides it in the Via: header.

I can tell you from a days logs I have handy, what it looked like for
the rest, though. From 1,680 different IP address/User-agents "via
..." combinations, you get 90 different version strings, and the top
20 look like:

  11 Squid Cache version 1.0.15"
  11 proxy gateway Spaghetti 1.51a"
  12 Squid Cache version 1.0.10"
  12 Squid Cache version 1.0.9"
  12 WBI Proxy Server 2.0)"
  13 Harvest Cache version 3.0pl5-Solaris"
  13 proxy gateway CERN-HTTPD/3.0pre6 libwww/2.16beta"
  16 Squid Cache version 1.0.12"
  16 Squid Cache version 1.0.16"
  17 Squid Cache version 1.0.beta17"
  18 Squid Cache version 1.0.5"
  19 Squid Cache version 1.0.17"
  24 NetCache version 3.1.1d-Solaris"
  26 Squid Cache version 1.0.18"
  29 Harvest Cache version 1.4pl2"
  65 Squid Cache version 1.0.22"
  68 Squid Cache version 1.0.20"
  78 Harvest Cache version 1.4pl3"
  82 proxy gateway WebTrack-HTTPP/1.2 libwww/2.17"
 957 proxy gateway CERN-HTTPD/3.0 libwww/2.17"

So, sadly, 957 different sites are still using the CERN reference
implementation. *Remember*, this list does not include Squid 1.1
sites, as I've not been logging Via: headers (well, it did have a 1.1
alpha 15 site, which is also pretty bad.)

>offhand, I can only think of 2 non-linux/freebsd caches, the
>cache farm (which I think is one of the biggest in the world, Sparc/Solaris),
>and's cluster of DEC's.. Both using squid.

I would think the largest probably include those in Singapore, which I
believe are Netcache on DEC

>We will only support operating systems that have a hostid function, so
>we can licence it. However, we haven't realised that we're throwing away
>80% of our market by doing this.

(they don't lock licenses to hostid.)

Being realistic, (and because we talk to the Netapp people quite a
lot), I think their main reasons are that it's very hard to spread
your development effort across a lot of platforms (they used to offer
a FreeBSD version, back in the IMC days).

The OSs they do currently support offer things like asynchronous I/O
(and however much we like FreeBSD, and Linux, they don't yet have this
sort of thing), and have a consistent set of libraries to develop to
(Linux is especially bad on this one.)

But obviously if you tell their marketting/sales people that you'd
really like to use it on FreeBSD or Linux they will listen and feed
that back to the product team and those making the decisions.

>Whilst the features of the software look -really- good, I'm not going
>to buy a $40,000 machine to replace my intel-powered $8,000 machine just
>so I can have persistant connections and a fast rebuild. (Not that I

As I think I've said before, buying commercial products isn't for
everyone. But some people find that it's more important to offer a
good solid service than spending a lot of time and effort doing it

-- jrg.
Received on Tue Sep 23 1997 - 07:37:01 MDT

This archive was generated by hypermail pre-2.1.9 : Tue Dec 09 2003 - 16:37:08 MST