Call it a Cache-22. Dynamic Web content is emerging as a key to successful e-business sites. But that fluid information can lead to heavy network traffic, slowing application servers to a crawl and, in turn, bringing an e-business to its knees.
Enter ESI (Edge Side Includes), a new caching standard that enables developers to craft applications that lower costs and minimize traffic congestion and customer response time. Products and services that support ESI are beginning to enter the market, and some enterprises are trying them out.
The move to display content that can change by the minute stems from the need to offer customers accurate information from product catalogs, auctions and online exchanges, as well as the latest stock quotes, news and weather. This ever-changing information is a far cry from the static pages and banner advertisements typical of early e-commerce sites.
Because much of this real-time data is on central servers, the pipes leading to those servers have become network traffic bottlenecks. Whats more, the servers are not the most cost- effective place to do all the processing. “Companies would like to offload processing chores from expensive central servers to inexpensive commodity hardware, but that hasnt been possible with dynamic content,” said John Magee, senior director for Oracle9ias product management at Oracle Corp., in Redwood Shores, Calif.
Offloading dynamic content can be tricky because its hard to know when some information will be updated or which user will request it. However, the vendors sponsoring ESI recognized that while a larger percentage of information was dynamic, there still were many similarities in the pages shipped to users. Therefore, they wanted to create an algorithm that would not ship every bit of information but would cache pieces that were repetitive to reduce network traffic. They also needed to divvy up Web page generation chores and then develop a way to deliver rapidly changing content to users based on different criteria such as the time of day.
Designed to function like HTML, ESI differentiates information that only a central server can generate from information that can be stored on other systems, which can be stationed in a companys data center or along the network edge as part of a private network or CDN (content delivery network). ESI then provides programming calls so the distributed items, called content fragments, can be aggregated, assembled and delivered as complete Web pages to end users.
The standard enables a developer to create dynamic content via templates that break down Web pages into HTML fragments, basically subsections of a page. Each template features common elements, such as logos or navigation bars, and is associated with a URL that an end user requests. The fragments are markup files containing discrete text or other objects, and each has a particular profile. For example, a company may want to store a template for several days but copy a particular item (say, a stock quote) every 15 minutes. In addition, fragments, such as a users bank account number, may be declared too sensitive to store locally and be kept only in a central system.
In addition to reducing hardware costs, spreading processing chores among a series of servers can improve the scalability, reliability and fault tolerance of an IT infrastructure. Whats more, the systems are more secure because an attacker has to shut down several servers, rather than one, to knock a companys Web systems offline, according to Peter Christy, founder of NetsEdge Research Group, a Los Gatos, Calif., market research company.
A Standard Emerges
Development of the specification stemmed from research projects at Akamai Technologies Inc., in Cambridge, Mass., and Oracle. Both companies were examining ways to improve delivery of dynamic information, although they had different starting points.
Akamai wanted to make its EdgeSuite CDN services more attractive, and Oracle was trying to enhance Oracle 9ias, its Web application server. “Since our customers work with a wide variety of products, we didnt see much value in trying to develop a proprietary technique for pushing dynamic content off of central servers,” said Thorsten Ganz, director of marketing at Akamai.
Unlike many standards efforts, this one emerged without a fight. “I dont know if the slowing economy played a role, but whatever the reason, no vendor thought it was worthwhile to develop an alternative to ESI,” said NetsEdges Christy.
After deciding to work together, Akamai and Oracle solicited help from other vendors, resulting in a group of 15 crafting the first version of the specification. The participants represent a variety of products: Web application servers from BEA Systems Inc., in San Francisco, and IBM, in Armonk, N.Y.; CDN services from Digital Island Inc., in San Francisco, and Mirror Image Inc., in Woburn, Mass.; development tools from Macromedia Inc., in San Francisco, and SilverStream Software Inc., of Billerica, Mass.; and applications from Art Technology Group Inc., of Cambridge, and Open Market Inc., of Burlington, Mass.
In June, the group completed the first release and then persuaded the World Wide Web Consortium to take over responsibility. In September, the W3C passed ESI as a note, its version of a standard. In addition, the supporters developed ESI for Java, extensions to the Java programming language that enable JSPs (JavaServer Pages) to call ESI components. JSPs are server-side software modules that produce a final user interface by linking dynamic content with static HTML through tags.
With that foundation laid, the issue then became persuading vendors to incorporate ESI support into products. “Adding new features to current products usually takes six to 12 months, so I expect the number of vendors delivering ESI-compliant products to increase from the summer through the end of the year,” said Neal Goldman, an analyst at The Yankee Group, a Boston market research company. Not surprisingly, Akamai and Oracle are leading the pack in that effort. In December, Akamai added ESI to its EdgeSuite product line, which pushes dynamic content to the network edge, and in January, Oracle included support for the interface in Oracle9ias.
ESI and the Times
The Seattle Times new media department turned to ESI after having trouble processing Web page requests when noteworthy news events occurred, such as the World Trade Organization meeting, an earthquake or the Sept. 11 terrorist attacks. “We werent sure how many hits were occurring because many individuals logged off since response time was so slow or they were denied access to a particular page,” said John Soltys, senior Internet engineer at the publishing company, which has about a dozen Sun Microsystems Inc. servers running its applications.
To solve the problem, the company deployed Akamais EdgeSuite services in February 2001 and added ESI functions in January 2002. There were a few hiccups at first. “Initially, we were missing information in the fragment headers, so it wasnt clear to the different systems what items had to be pulled from our Sun servers and what data was stored in the Akamai network,” said Soltys, adding, “We worked with Akamai to resolve that problem.”
Akamais implementation of ESI is different from that of Oracle. While the Akamai implementation pushes content off the central server and on to the network, Oracles pushes the content from a database server to an application server.
Macromedias implementation, meanwhile, moves content to client systems. “Akamai has focused on using ESI to push processing functions onto its network; we think there are instances when companies will want to move it back onto the client system,” said Kevin Lynch, chief software architect at Macromedia.
When more ESI-compliant products ship later this year, companies should gain flexibility in designing applications as vendors implement ESI in different ways.
Whatever the scheme, the goal remains the same: to get data off a central server to lower costs and improve performance. That mission is not a little ironic, some observers say. Application architects expected many benefits from Web-browser-based applications, not the least of which would be improved manageability, thanks to moving processing from client systems to central servers.
Paul Korzeniowski, a freelance writer in Sudbury, Mass., specializes in networking issues. He can be contacted at [email protected]