Rohit Khare (
Fri, 23 Jan 1998 17:34:56 -0800

$40k box with honkin' disks acts as a serious web cache, with active
prefetching and an embedded OS tuned for the task:

I wonder how much of http/1.1 caching logic *they* support :-) They go one
better by actually studying the change frequency and realizing which files are
logos (change never), ads (change hourly), and so on -- as well as likely link
traversal frequency for prefetch.

Gee I wonder where their W3C membership check is? Seriously, what would they
expect from being there? They have a product to ship now, like Sitara; but on
the other hand, it's the NG committees that have the most interest in getting
experience info *from* these pioneers.


As they see it:

Active Caching
Predictive prefetch and adaptive, asynchronous refresh of Web

Optimized Storage System
Minimize fragmentation of saturated disks for fast reads and writes

Smart Bandwidth Utilization
Refresh Web objects using off-peak bandwidth, and reduce peak-hour
HTTP traffic

Easy Administration
Simple installation, high reliability, self-optimizing adaptive
algorithms, and secure browser-based management

Active Caching
With CacheOS, Web pages are actively cached on a per-object basis.
When a Web
page is requested by an end user, CacheOS tracks a variety of
information for each
cacheable object in the page (e.g., text, images, Java applets,
sound). Variables that
are tracked include frequency of request, frequency of object
modification at the
server, and time cost to retrieve the object. CacheOS then
automatically determines
the refresh pattern appropriate to each object that is likely to be
requested again. This
consistently provides updated, complete pages from the cache an
order of magnitude
faster than from other Web caches.

Object Prefetching
As an Active Cache, CacheOS also prefetches Web objects. When a
client requests a
Web page that is not in cache, CacheOS uses patent-pending
Predictive Prefetch to
retrieve the objects from the Web site faster than the end user's
client can request
them. The objects are then delivered to the end user's screen at
LAN speed,
significantly reducing the delay for first-time page hits.

DNS Caching
CacheOS maintains a large DNS cache that is refreshed in a manner
similar to the way
Web objects are actively kept fresh. Since DNS lookups can
contribute five or more
seconds to total end user delay, this shared DNS cache gives a
significant boost to
overall Web cache performance.

Optimized Storage System
Caching is unlike any other application. The disks of a Web cache
should run almost
completely full all of the time, since full disks increase cache
utilization. In addition,
most objects on disk are subject to removal at any time in order to
make room for new
incoming objects. Consequently, it is imperative that a Web cache
be based upon a
storage system optimized for caching.