Caching: How Real the Benefits?

Caching technology has finally established a firm foothold in Internet access applications, but the debate still rages over how beneficial it is—and for whom.

by Jim Thompson

The promise of increased speed/decreased latency without the cost of additional bandwidth has catapulted caching software and appliances into a prominent place among the fastest growing segments of Internet technology. At the same time, there's a good deal of skepticism and conflicting opinion about these products.

Broad claims for new technologies often turn out to be exaggerated—or at least over-generalized. While there's little doubt, in theory, that caching should benefit Internet-access performance, it's far from being accepted as gospel in the ISP community that cache is a practical, cost-effective investment. Sure, AOL got itself past some dark days by implementing monster Web caches, but does the technology really make sense for more typical ISPs?

Bandwidth or cache?
A common perception that militates against cache solutions is that it's more effective simply to add bandwidth. As the cost of bandwidth falls, many ISPs believe this solution to be the more cost efficient approach to increasing performance.

Those who have actually been running caches, however, disagree that bandwidth is necessarily an effective remedy for the Internet bottleneck. "That's horse feathers!" said George Peabody, analyst with the Aberdeen Group. Peabody explained that as access bandwidth in the local loop increases, the need for caching also increases. "For example, it's important that streaming media—video and audio—be as close as possible to your customers, and caching is the way to do it."

Joel Yaffe, analyst with the Giga Information Group agrees. He notes that bandwidth (and its cost) can "actually be a constraining factor," since it doesn't address many of the inherent performance problems of the Internet. "By caching content closer to your users, the requests don't have to go out over the Internet—a benefit that's not tied to increasing bandwidth," said Yaffe.

Mike Gibbs, director of engineering and operations for Virginia-based ISP and backbone provider ServInt Internet Services, has been running a cache for several years. In his view, adding an additional T-1 line will not solve the latency problem for the small to mid-size ISP. "Most ISPs over-saturate or over-sell their lines. This is the only way they can make money. Adding a second T-1 only means that it, too, will be over-sold." He added that with a cache behind the network, the ISP garners a double benefit. "First they can over-sell even more and, second, their customers won't suffer from that over-selling as often as if there was no buffer in between."

"Adding bandwidth doesn't eliminate the bottleneck, it just moves it," Gibbs pointed out. Without caching, a server has to send data individually to anyone wanting content from a specific Web site. In a white paper for Inktomi Corp., Michael A. Goulde equates this to "everyone having to fly to Hollywood to see the latest movie" instead of sending copies of the film to theaters across the country where it can be viewed by thousands at each location. Congestion on the Net adds to the latency and decreases the overall quality of service for all users.

Payoff points
Caching critics argue that it's a viable and cost effective solution only for the largest service and content providers. "Not so," states ServInt's Gibbs, who adds that in his experience a cache is more usable and economical for the small to mid-sized ISP than for a larger ISP like UUNet. "A cache provides a speed increase with a saving of bandwidth, which the big players, in theory, already have. The smaller ISPs who are thinking about upgrading their connectivity can employ a cache to resolve some of their speed or congestion problems."

But some disagree. "Caching for the smallest ISPs has generally not been a pay-back situation," said Peabody. He notes that one of the arguments for caching is that it can be an alternative to or a replacement for a T-1 line. "An ISP would be better off with a T-1 from another provider, so there are two routes out. Plus, there must be a significant number of hits to populate the cache. For smaller ISPs, the numbers simply have not been enough to warrant caching."

Peabody added that caching "starts to make a lot of sense" for the smaller ISP's upstream providers and for those tapping into the global marketplace.

Running the numbers
The question of whether or not caching is a cost effective solution is one that requires careful calculations based on usage patterns and an organization's network configuration. "The calculation involves the amount of cachable content, the cost of the cache, and the cost of maintaining it. This has to be weighed against potential benefits to the users," said Yaffe.

On the cost issue, Peabody underscored the fact that bandwidth is not only expensive, it's also a recurring cost. "With caching software or a caching appliance, you buy it once and put it in place," he said.

"Caching is no longer an expensive solution," said Gibbs who notes that a number of low-cost caches are now available. "CacheFlow and InfoLibria, for example, have affordable 'pizza box' solutions that are designed for T-1 or dual T-1 clients. Some sell for under $10,000, some are under $5,000, but all provide significant benefits.

"We have hit rates of anywhere from 40 to 60 percent off of our caching device," said Jerry Kalligonis, director of information technology for Pennsylvania-based ISP BlazeNet. "At those rates, there's no question that the benefits outweigh the cost."

It seems clear then that caching is a viable performance-enhancing solution across a wide range of ISP situations. The question ISPs must ask is whether or not it's cost efficient in their particular case. "Caching certainly makes sense. The question is how much do you save and, is it worth the cost of devoting resources to that function," said Giga's Yaffe.

—End

Read Caching 101: What's the Return on Investment?