InformationWeek.com

Welcome Guest. | Log In| Register | Membership Benefits

June 21, 1999

Print this story

No Need To Panic--Yet

continued...page 2 of 3

Related links:

For more on dedicated E-commerce servers, see our product comparison

"If you build the infrastructure yourself, architect the site for speed and reliability," King says. There should be no single point of failure, and the infrastructure should be modular to accommodate growing traffic volume and new types of users. Barnesandnoble.com collocates the site in three places. If a crash occurs at one site, the other two would seamlessly handle the load until a fix is found. As a further safeguard, redundant server farms at all three sites provide seamless local failover.

Collocation can help with traffic congestion, where server load-balancing devices route user requests from busy servers to idle ones. Some backbone vendors offer collocation services as an add-on to high-speed backbones.

Collocating the site's architecture wasn't the most challenging aspect of the site, King says. Building an E-commerce platform and tying it to Barnes & Noble's existing systems caused his biggest headache. The company built its own software because off-the-shelf supply-chain and enterprise resource planning software didn't work. In traditional retail, King says, you buy, sell, and store products. On the Web, you sell, source, and deliver them, so different methods were needed.

Barnesandnoble. com's effort began in 1996 and the Web operations were launched in May 1997. Since then, the site has gone through several iterations, culminating in the present infrastructure. It consists of three basic server farms covering search, content, and transaction tasks. All are based on Hewlett-Packard Lx Pro servers running quad 200-MHz Pentium processors with 2 Gbytes of RAM and an average of 12 Gbytes of RAID 5 hard disks.

Many of King's hardware and platform decisions almost made themselves; it may be the same for you. For example, whether to code your own commerce server application or rely on a third-party package such as Microsoft Site Server may be resolved by your Web development team. If these decisions are up to IT, managers should first consider back-end legacy requirements and existing operating-system preferences.

For instance, Windows NT and Microsoft Site Server are excellent choices in terms of price and horizontal scalability. But it's critical to consider the specific network. If a company uses IBM RS/6000 or AS/400 servers, it may use Lotus Domino as the workflow agent and IBM's DB2 at the back end. Adhering to single-vendor product lines not only speeds integration but may also let managers negotiate more favorable purchasing contracts.

Traffic Jams
Part of choosing an appropriate server platform is knowing what kinds of traffic loads are expected on back-end legacy systems. Large-scale efforts, such as extranets that involve back-end network resource sharing, can add significant load and stress on the network as a whole. When business partners are given direct access to an internal inventory database for automated supply-chain management, database performance has to be up to the task. Also, the system and the rest of the network have to be protected from outside threats, such as hackers, viruses, or unauthorized applications.

Secure E-business must take into account security's impact on network performance. Hardware firewalls have traffic limits that need to be accounted for. If throughput is too slow, managers should consider multiple firewalls and a load-balancing device to route requests to idle servers after security verification.

The most insidious security problem is behind the firewall at the transaction server level, where network servers encounter encryption technology that depends on calculations that cut transaction throughput. There are two ways to address this problem: software architecture that avoids encryption unless absolutely necessary, or hardware crypto-accelerators.