CACHE VALUES

Got a bandwidth crunch? Throw cache at it.

That's certainly the impression a service provider can get from the reports flowing from the growing ranks of cache vendors.

Everyone knows the conditions that have focused attention on caching. For one thing, there's traffic flow: Exponential growth in the number of Internet users has more people demanding more of the content that already exists on the Web.

For another, that content constantly is chewing up more bandwidth as the Web becomes a force to be reckoned with in news, multimedia, Java applets and flashy graphics. Every time a space shuttle launches or a breaking news event hits the Web, caching keeps the bits moving to the millions who want them. Finally, e-commerce in all its forms is placing new strains on the Internet and provider networks as more users log on to perform regular tasks such as shopping, bill payment and on-line investing. And "always on" technologies such as cable modem and DSL technologies are bound to lead the surfing public to adopt "always available" expectations for Web content.

But the decision to opt for caching carries a trail of secondary choices that will determine a network's shape, performance and operating overhead. Most service providers find that just saving bandwidth isn't enough to justify the expense of deploying caches.Instead, they are more interested in the advanced applications that caching can allow them to do.

"The folks who do business hosting are much more interested in putting content-based services on their networks than they are in the one-off benefits of caching alone," says Kevin Lewis, marketing director for InfoLibria. "Things like distributed hosting, high-quality media services, high-quality or customized ad insertion - that's where we think the service providers will end up in terms of how they value caching."

LAYING OUT CACHE

Caching involves storing frequently accessed content closer to users so they have to cross fewer networks to reach it. The user's browser sends out a request for Web content or data over the Internet, and the request is directed to the network caching server located somewhere between the browser and the originating Web page - often at the user's ISP or some other point of presence on the Internet. If the server has a fresh version of the content, it sends that content to the end user. If not, the server goes back upstream to retrieve the content from the proper Internet source, fills the browser request and usually stores a copy of the content in anticipation of future requests.

Software-based caching runs on a general-purpose proxy server; caching appliances run caching software on a dedicated hardware platform. The former essentially is an option only for enterprises, located behind the firewall and caching commonly accessed Web pages for users in the LAN. Caching appliances operate at higher speeds, cache more efficiently and are thus the carrier-grade choice.

Caching appliances can work in one of two ways: by proxy caching and by transparent Web caching. Proxy caches are located explicitly on the public Internet infrastructure and act as proxy agents for the browser. Instead of contacting providers' Web servers, browsers send their HTTP requests to the proxy, which fulfills them if the content is present on its disks, or refers back to the originating server if not.

Because each client browser must be manually configured to send Web requests to the correct cache, this setup is very labor-intensive. And because requests generally cannot travel from one proxy cache to another, each cache must maintain a copy of all the cached material it might need to supply, making availability a concern.

Transparent Web caching - in which the cache still acts as a proxy for the browser but now sits in the data stream - eliminates many of these problems, most notably the burden of configuring browsers. Transparent caching can be implemented in several ways. Most basically, it can be done by placing the cache directly in the data stream, which requires routers or load balancers to direct traffic into the cache.

Most cache products support clustering, with groups of caches either at the same site or geographically distributed talking to each other via standard protocols. Two primary standards permit cache clustering: Internet caching protocol (ICP) and cache array routing protocol (CARP). Many products support both. Another, Web cache control protocol, is supported by most of the appliance-based cache manufacturers.

One important consideration is that clustering protocols can reduce speeds. ICP checks to see if an object is available in any other cache, thus reducing its performance speed. CARP doesn't check all caches, but it still can't forward a request at the speed of the standard IP switch.

Lucent Technologies recently entered the caching field when it introduced its IPWorX platform, a distributed caching product that groups caches around an intelligent switch. The Layer 7 switch, called WebDirector, uses a hardware-based forwarding engine to switch flows between Web server farms and users on the Internet, distributed caches, or in corporate intranets or extranets. It identifies traffic by criteria ranging from IP addresses to URLs to "cookies" - information on an end user's computer that links it to the server hosting a specific Web site. That permits the engine to know the type and location of all network content and to direct server requests more efficiently.

ACTIVELY SEEKING CONTENT

Passive caching checks the content server every time a request is made to learn whether content has changed and retrieves a new copy if the content has been modified. This method is slow, because every request is checked before being served.

A more efficient method is active caching, in which changeable sites are updated more frequently than relatively static ones - either on a scheduled basis or automatically.

"Fast access to stale content is no service improvement," says Kelly Herrell, vice president of marketing for CacheFlow.

Most caches perform synchronous downloads - that is, they make the user go back and get each item from a Web page. CacheFlow speeds the process, says Herrell, by modeling individual objects on a Web page, sometimes as many as 40 per page. These change at different rates, so CacheFlow's operating software builds a "change probability" profile for each object back at the server. The cache then adjusts to stop refreshing the CNN logo and spend more time pinging the server for the lead story, making it more likely that the content is the freshest available.

What kind of content can be cached? Anything written in HTTP, certainly. Most caching products today also can accommodate content written in file transfer protocol.

Some products such as Inktomi's Traffic Server and Entera's TeraNode caching appliance include a flexible, open-standards plug-in software architecture to allow caching and distribution of network news transfer protocol content, which includes news feeds and discussion groups such as Usenet. "It's not a sexy area or a big profit center for provider, but the Tier 1 players will tell you it's a big replication problem for them," says John Scharber, Entera cofounder and chief technical officer. "Back in the mid-'90s, they were pushing 2 or 3 GB of total news a day. They went from pushing 20 GB of news last year to 40 in the first six months of 1999."

Streaming media is one of the golden value-adds in the caching space. Inktomi was the first to solve the problem of caching this content with its release of Traffic Server 2.0 last summer in conjunction with Real Networks. That platform, since broadened in Traffic Server 3.0, is still one of the most tightly integrated caching solutions for streamed audio or video. Real Networks contributed software that permits a different codec for each downstream line speed, allowing the appliance to optimize for 28.8 kb/s up to ISDN or DSL speeds.

Live streamed events can't be cached, but Traffic Server can help. "When lots of people want to view the same live content, we can split that content stream out to tens or hundreds of clients," says Peter Galvin, director of Traffic Server for Inktomi. "This eliminates a lot of congestion on the Internet. When all those people logged on looking for Clinton's testimony, the number of people taking streams from the servers made the quality of the video very grainy in many cases."

InfoLibria combines its DynaCache with another product called Mediamall, which serves streaming audio and video content from the edge of the network.

Finally, because caches are located at the edges of the network - where many other functions are migrating - they are a natural spot for the convergence of other network management services. Manufacturers have begun to build numerous edge functionalities into their caching products. Many offer security products such as firewalls, and most offer URL filtering to keep users away from protected sites. Directory services have begun to appear and will continue to serve the growing demand for user authentication and remote sign-on.

The reverse is true, too, at least on the enterprise level. Firewall vendors and Internet server manufacturers such as Nortel are offering caching capability in several of their products aimed at small to mid-sized businesses.