Always There for You

E-commerce requires a highly available web architecture - and clustering may be the answer

By Kamesh Namuduri & Palani Ram

In today's global marketplace, your Web site has become indispensable for communications and transactions with suppliers, partners, and customers. Any slowdown or outage affects not only your company, but also your business partners and your bottom line. In this situation, your Web site must be highly available; your Web architecture must provide But how can you achieve 99.999 percent availability and other IT objectives without spending your entire budget? Even with technology rapidly changing and becoming obsolete, your computer systems must be useful as long as possible. As your business grows, your computer systems must grow as well - but you shouldn't have to replace a majority of them to do so. Instead, you want to easily manage and administer the existing systems alongside the new. Furthermore, working systems should hedge the failed systems to provide fault tolerance. The cost-effective solution to this IT challenge is clustering.

A cluster is a group of individual computer systems (usually referred to as nodes) that appear as a single coherent unit. In this column, we'll discusses some of the business drivers that are making Web server clustering or server farms necessary for today's business Web sites.

Web Site Dependence

A business Web site usually provides information about the company for its customers. E-commerce has dramatically changed the Web site from just an information-sharing bulletin board to a fully transactional business interaction with supply chain partners and customers. The Web site with static information has transformed into an intranet-, extranet-, and Internet-based customizable application media.

Your business enterprise uses the Web to generate revenue, satisfy customers, and optimize business processes. Your Web site is not isolated within the company or with your customers anymore. Now your Web site supports and depends on your business partners and their enterprise systems. For example, a furniture Web site that provides custom-designed furniture may query its supplier's inventory in order to provide the delivery date of a chair to a customer.

When your company first launched its Web site, your infrastructure could easily meet its current and near-future demand. The problem is that you can't predict how much that demand will increase. Although a growth in demand may eventually turn into revenue, the higher the demand, the more your company must scale its infrastructure to meet that demand.

Your Web infrastructure must respond quickly and always be available to your customers. One unfortunate effect of high demand is that response time increases and server availability decreases, which translates into turning away your customers and your partners' customers. When your Web infrastructure is unable to respond to increased demand, it not only affects the revenue and creditability of your company, but also your business partners.

Web Traffic Demands

Increased traffic into and out of your company's Web site may have many causes. For example, your company may decide to reduce paperwork and handle business transactions online - an increasingly common business-to-business (B2B) scenario. Other reasons for increased demand include the Web site's popularity either year-round or during the holiday season. High traffic is beneficial for your company only if you can meet this demand by scaling your Web site proportionately.

Your customers will visit your Web site only if the response times are short. On average, customers expect the site to download within eight seconds. When that response time goes up, you have given your customer the incentive to visit a site with faster response times. If you lose customers because of long response times, you are facing huge losses for your company as well.

When your Web site is down, it affects your customers who use the site regularly. In June 1999, a 22-hour outage at the eBay Web site cost the company $3.9 million in credits given to users. Outages diminish a company's revenue and reputation, which can make estimating the dollar amount of downtime very difficult. An excerpt from the eBay 1999 annual report on the outage states, "In addition to placing increased burdens on our engineering staff, these outages create a flood of user questions and complaints that must be responded to by our customer support personnel. Any unscheduled interruption in our service results in an immediate loss of revenues that can be substantial and may cause some users to switch to our competitors. If we experience frequent or persistent system failures, our reputation and brand could be permanently harmed."

In addition to your company's direct customers, an outage may affect your business partners' ability to complete transactions for their direct customers. These chain-of-business transactions may connect to several other business partners, in which case, the impact of not completing the transaction may have serious consequences. Thus, in addition to a loss of business, you may also incur damages from your business partners. Your company may lose its credibility because of the Web site's downtime.

Web Infrastructure

Web architecture comprises several subsystems such as firewalls, Web servers, transactional servers, application servers, database servers, and network infrastructure (Internet routers and Web switches). Each subsystem has its own limitations in handling the increase in Web traffic, which can slow down the performance of the Web site. The architecture's performance may decrease because of increased Web traffic as a result of either multiple points of failure or a single point of failure.

Multiple points of failure. This decrease in performance or unavailability of a system is due to the failure of two or more subsystems. For example, in a poorly designed Web-based application system, a seasonal increase in the Web traffic may overload the server. Furthermore, downstream systems such as the application servers and databases may lock up if they don't have enough memory and processing power, causing the systems to slow down significantly. The increased response time in this case is due to multiple failures. You should analyze your Web architecture for the increased traffic, and you should scale bottlenecked subsystems for the processing power to improve the overall performance of the Web architecture.

Single point of failure. Failure of any one component or subsystem may bring the whole system to a crashing halt. This can be any subsystem such as a router, Web server, application server, database server, or any particular software. An economical solution is to provide a backup mechanism for this single point of failure. The money invested in providing this redundancy is well justified by making the Web site continuously available to customers and business partners.

A Scalable Solution

Scalability is the ability to increase the processing of a system to accommodate increases in Web traffic and to respond to HTTP requests within a reasonable and satisfactory time period. A Web server may "spawn" multiple threads (lightweight processes) to respond to multiple HTTP requests simultaneously. However, as the number of threads increase, the response time of the Web server will also increase because these threads need to share CPU time.

One performance boosting solution is to scale the Web site by replacing the server with another server that has a faster processor or multiple processors. Although this solution may work in the beginning stages of business growth, the single server solution may not be as cost effective in the long run. Furthermore, neither of the servers may be available during the "replacement window." The alternate solution? Distribute the workload of incoming HTTP requests across several servers.

Rate This Article Rating: 1 (best) 2 3 4 5 (worst) Comments: Optional e-mail address:

A number of subsystems achieve the response to the client from the server for each client request, and any one of these subsystems can be a single point of failure. Clustering can solve these problems because many units within a system work together as one system. The outside of the system does not know about the failed unit, because the processing is directed to the healthier units within the cluster.

If you want to find out how to properly cluster your subsystems to eliminate single points of failure, stay tuned. In our next column we'll delve more into clustering and how to achieve scalability and high availability for your Web infrastructure. Ultimately, lost revenues from Web site unavailability outweigh the costs of Web infrastructure redundancy. Faster response times will encourage more businesses to operate through the Web and decrease the time spent by individual employees on business transactions. And this in turn reduces the cost of doing business on the Web.

Kamesh Namuduri (kamesh@mediaone.net) is an assistant professor in the computer and information science department at Clark Atlanta University in Atlanta.

Palani Ram (prr@webahn.com) is an IT consultant. He currently designs Web-based enterprise architectures and B2B systems.