Welcome Guest. | Log In| Register | Membership Benefits

News

June 5, 2000

Printer ready
Printer ready

Three Keys For Building A Super-Fast Web Site

By Jason Levitt and Logan Harbaugh

Illustration by Campbell Laird A Web site today is rarely a single machine that serves up pages. Increasingly, Web sites represent distributed areas of network intelligence, Web content, and application services across networks to optimize Web response time. Building a super-fast Web site can be boiled down to three architectural issues: caching, routing, and load balancing. Web administrators must use a combination of these hardware and software techniques to optimize performance.

Caching is the act of moving data--graphic files, HTML code, or dynamic content--around the network to best meet the needs of network users. Databases, for instance, are always cached. But caching the results of database queries for Web transactions is a complex task that no vendor has sufficiently solved. Static and dynamic Web pages can also be cached.

Typically, static Web pages are served from one set of machines, while dynamic content, which usually includes database data, comes from another set of machines. It's possible to cache static Web pages almost completely in RAM so that they get served ultra-fast.

With dynamic pages, Web sites can use predictive algorithms to help answer questions such as, "What did the user want last time?" or "What do most users usually want at this point?" This approach allows dynamic pages to be generated before the user actually requests them.

Routing user requests to the closest data center can be done with software or service offerings. Basically, a large site will have some or all of its Web pages replicated at different data centers around the world; when users request pages, they get them from the site that can deliver them the quickest. This is an important method used by all serious Web sites that serve a global audience and need to have fast response times.

One major technique that has evolved for ensuring Web-site performance is load balancing. This lets the administrator set up several Web servers and share the load among them, keeping any one server from being overloaded. It also provides fault tolerance, since a single server failure doesn't affect the entire site.

As vendors strive to differentiate themselves in this market, new products with new features are being developed. They include the ability to have several servers for each of the major parts of a Web site, geographic load balancing, and Secure Sockets Layer encryption accelerators that offload the encryption processing from the Web server to a dedicated appliance.

Another technique for speeding up a Web site involves having multiple, redundant Internet service provider connections to the Internet. Not only does this ensure that an ISP problem can't take the site down, it allows for geographic load balancing, where visitors are sent to the connection closest to their location.

Once a Web site has multiple, high-speed connections and a server farm, the next bottleneck becomes the back-end databases that provide content for the site. Issues here involve connections to the database server and performance of the server. Connections are being upgraded to Gigabit Ethernet, which should be sufficient for even large Web-farm front ends.

The last bottleneck is the applications themselves. A number of large sites don't properly debug their code, meaning that recovering from a crashed browser can take longer than waiting for a slow connection.

Return to main story, "Fast Is No Longer Fast Enough."

Illustration by Campbell Laird

Back to This Week's Issue
Send Us Your Feedback
Top of the Page