Cache cachet
Web caching appliances help conserve precious
Internet bandwidth.
By NEAL WEINBERG
Network World, 05/10/99
Is your Internet connection getting
bogged down? Are end users
complaining about the World Wide Wait? Are you
adding new Internet or intranet applications that may
require additional bandwidth?
Maybe it's time to consider implementing a
Web-caching appliance. These devices store a copy
of previously accessed Web pages and eliminate the
need to download those pages from the Internet the
next time an end user requests them. By reducing
Internet traffic and improving network performance,
caching appliances may allow you to delay a costly
Internet access upgrade.
ISPs buy the bulk of caching products these days, but
analysts predict that corporate customers will begin
turning to caching appliances later this year as they
either outgrow their software-based proxy servers or
look to caching for the first time in the face of
mounting end-user complaints about slow Web
response time.
"The market for caching will ultimately be huge,'' says
Peter Christy, an analyst at Collaborative Research in
Los Altos, Calif. He expects the market to grow from
$100 million in 1998 to $2 billion by 2002, with
caching appliances dominating the field.
However, caching appliances are expensive - a
high-end box can cost $100,000. And early adopters
say that while caching helped them reduce Internet
traffic, they still had to upgrade their Internet
connections to accommodate exploding user demand.
"Caching is not magic," Christy says. "It doesn't mean
all you'll ever need is a T-1 line.''
Ray Smith, Internet services manager at IT consulting
firm Computer Sciences Corp. (CSC), knows that
only too well. Workers at CSC's 450 remote offices
connect over the company's WAN to its central office
in Falls Church, Va. From there, they access the
Internet via three separate ISP connections that
provide 9M bit/sec of total bandwidth.
However, bandwidth requests on the WAN were
peaking at 18M bit/sec, so the Web traffic was
dragging down the entire remote access network.
Smith deployed two CacheFlow appliances from
CacheFlow, along with an ACEdirector Layer 4
switch from Alteon WebSystems. The switch
intercepts all HTTP traffic and redirects it to the
cache, reducing the load on the company's remote
access router and improving overall WAN
performance.
The appliances also reduced the amount of Internet
traffic, and end users noticed a quicker response to
their requests for Web pages. The CacheFlow
products were reporting an impressive 65% hit rate,
which refers to the amount of requested material found
in the cache. Vendors put the average hit rate at about
35% to 40%.
But to Smith's chagrin, "the bandwidth we saved
quickly got eaten up by something else." The
company's 2,000 consultants began to more heavily
rely on the remote access links for other applications,
and their appetite for Web surfing also increased. As a
result, he's been forced to upgrade his Internet access
to 32M bit/sec with a variety of fractional T-3
connections.
Still, Smith says he loves everything about the
Cache-Flow appliance, with the exception of its
yellow color.
Jeff Wade, network information specialist at Science
Applications International Corp. (SAIC), a systems
integrator based in San Diego, has a similar tale.
About two years ago, he installed Netscape's proxy
server, which starts at $525 for a 100-user license.
Proxy server software is inexpensive compared with
caching appliances, but it can't touch the performance,
management features and scalability of boxes that are
fully optimized for nothing but caching.
Proxy servers have another drawback - to get the
system up and running, end users need to point their
browsers to the proxy server, which can cause an
administrative nightmare. Managers need to
reconfigure each browser to send all HTTP requests
directly to the proxy server's URL.
Wade supports 25,000 end users at more than 150
locations across the U.S. and doesn't have a way to
remotely change the settings on the end user's PCs or
send IS staff to all the locations.
He sent out notices through various internal channels,
but after six months, only 30% of end users had
complied with requests to reconfigure their browsers.
That meant 70% of end users weren't going through
the cache.
Looking for improved performance and better
management features, Wade then invested $60,000 in
a NetCache appliance from Network Appliance. That
price tag includes 100G bytes of disk space and a
redundant power supply.
He set the policy on his Cisco routers so that all HTTP
traffic is automatically funneled through the cache. This
allows the NetCache appliance to run in what is
known as transparency mode, which eliminates a
major administrative burden by removing the end user
from the equation.
Wade is getting a hit ratio of about 50%, but a more
important measure of bandwidth utilization - actual
byte savings - is only about 28%. He explains that the
cache contains a lot of icons, images, logos and tiny
dots that don't change, so the hit rate for these tiny
.GIF files is extremely high. However, the content that
needs to be refreshed and downloaded more often
includes large data files.
Nevertheless, Wade is pleased with NetCache's
overall performance. He's getting two million hits per
day and has more than two million objects stored in
the cache. "The machine doesn't breathe hard at all,"
he says.
The caching devices allowed him to forestall an
Internet access upgrade for several months, but
galloping bandwidth demands forced him to upgrade
his 4.5M bit/sec pipe to 10M bit/sec a few months
ago.
There must be a cache-22
Cache appliances have their limitations. First, pages
with cookies and dynamic pages, such as those
generated by Common Gateway Interface scripts, are
uncachable, although the vendors claim that those only
amount to around 5% of requests.
In addition, caching only applies to HTTP traffic, not
to other types of traffic that move over Internet links,
such as e-mail and File Transfer Protocol. Also,
caching doesn't work with pages that are generated
through a request made to a database or data
warehouse.
Vendors with software-only proxy server products
include Netscape, Novell and Microsoft. Caching
appliance manufacturers include start-ups CacheFlow,
InfoLibria and Network Appliance, which offer boxes
that run on proprietary operating systems.
Cisco's caching appliance, Cache Engine, uses a
proprietary operating system that works best when it's
connected to a Cisco router running IOS. And
Inktomi is targeting its caching software for SPARC
servers at carrier-class ISPs.
James Staton, an analyst at Dataquest/ Gartner Group,
says caching appliances will come to dominate the
market because of their superior performance. ISPs
will be buying most of the caching appliances in 1999,
but corporate interest will begin to perk up later in
1999.
It all depends on when network professionals begin
feeling the pain of user complaints over sluggish
performance, Staton says. He suggests network
managers start with a large cache at the data center,
then consider caching at satellite offices in other
countries, because international connections are so
expensive. Finally, move to caching at domestic
branch offices.
If you have a popular Web site, Staton recommends
implementing something called reverse caching. This
entails placing a caching appliance at the Web server
so it can deliver frequently requested pages to Internet
shoppers. This has the double benefit of offering better
performance to potential customers and taking some
of the load off the Web server.
Christy adds that the market for Web caching
appliances can only grow as companies begin to run
more Web-based applications.
He says the bottom line is that disk space costs less
than WAN bandwidth, so it makes more sense to
store copies of information locally than to go out over
the WAN and retrieve it. |
