It's The Server, Stupid

If e-commerce is to flourish on the Web and deliver on its promise for customers andvendors -- both outside and inside an organization -- there needs to be a mechanism toensure that what happened to Anwar Murbarat doesn't happen to you, or to me, or to anyone.Ever. In Murbarat's case, rather than prowling the local brick-and-mortar dealershipstrying to buy 20 new cars for his new cab company in Chicago, he tried Web shopping. Aftera few searches and inquiries he found an auto wholesaling market in Atlanta with a greatdeal. So far, so good!

As he started to execute the order for the 20 cabs, the auto dealer's Web site sloweddown so much that he got tired of waiting for a response. Frustrated, he tried again thenext day. Again, his efforts met with the same response. Finally, Murbarat called thedealer (whom he would not identify) and discovered that the Web site, overwhelmed withqueries about Chevy Malibus (a staple vehicle in the taxi trade), had lost his order inthe online chaos. Annoyed by this kind of poor service, he took his business --and tens ofthousands of dollars -- elsewhere.

Surf And Shop Stop

But what disappointed surfers and e-shoppers like Murbarat don't realize (nor shouldthey) is that the Web or what's behind the Web -- the packet switching foundation uponwhich the Internet rests -- is a first-come, first served technology. Unfortunately, mostreal-world business models are hardly so egalitarian: We reward frequent flyers withspecial deals, important shoppers with early notice of sales and high rollers with freetrips to Las Vegas.

And this kind of special treatment is not limited to e-commerce shoppers. Inside mostcorporations, as we already know, there is a natural pecking order of servicerequirements: Corporate executives need priority access to information to make strategicbusiness decisions. And we all know that in practice, some employees are more equal thanothers.

When a Web server is overwhelmed with requests, increasing transmission speeds to andfrom the server doesn't help the server process the requests. In fact, it may only"serve" to make matters worse. As the weakest link in the Web, if the Web serveris overloaded, improvements in the other links do not matter.

But if the Internet's democratic model doesn't fit, how can businesses ensure thatestablished, real-world relationships are reflected on their Web sites. Or in the words ofLisa Nash, writing for the San Jose Mercury News about the well-publicized E*Trade outagesback in February of this year, "The expectation is for real time."

"Internet Service Providers are building out plenty of bandwidth. [But] when usersconverge on a single server, the server becomes the bottleneck, not the network,"says Jim Zepp, HP's product manager for HP Quality of Service Technology. "Serverlatency [the time an Internet packet goes into the server and comes out] is so muchgreater than network latency [the round trip response time for a packet] that we think theserver is the biggest challenge."

Vive La Differentiated

By providing "differentiated services," Internet software and routingequipment can effectively mark IP packets with special flags that provide for differentlevels of service. Just like a "priority stamp" on a mail package is anindication to give the package special handling, marking packets requires bilateralagreements between those marking the packets and those forwarding the packets. Theseagreements, resulting in better service between customers and service providers, ensurethat packets can be classified for different levels of service on a "hop-by-hop"basis.

Differentiated services and server latency are at the core of HP's Web Quality ofService (WebQoS) software. First introduced in May 1998 (and first implemented in HP'sinfrastructure for the 1998 World Cup Games), WebQoS 2.0 (introduced in March 1999), isserver-based software that fits at the TCP/IP socket level and provides access andperformance priorities via the Request Controller. A separate component, called theResource Controller, sits above the OS layer, occupies a level below a company'smiddleware software and manages the resources allocated to the Web server.

"We're monitoring based on service objectives configured through the managementsystem and you can [allocate] how much of your resources are going to whateverapplications and which have what priority," explains Zepp. Interestingly, hestresses, "We can set the type of 'service bit' on request that comes out."

According to a recent analysis by the Giga Information Group (Cambridge, Mass.), oneU.S.-based manufacturing corporation was recently set to upgrade its high speedcommunications lines at several of its remote sites suffering from Internet trafficcongestion. By prioritizing Web-based traffic instead, the company was able to postponethe costs of the upgrade and immediately recover the costs of the QoS implementation.

It's similar says Zepp, to a car pool lane. "You're reserving capacity forcustomers who have two people or more in the car."

HP WebQoS software provides new services which are built with the Internet enhancementsin HP-UX 11.0 in mind, but it works with 10.20 as well. And although HP's WebQoS 2.0 wasdesigned to be Web-server independent, Zepp emphasizes that, "We've focused ourdevelopment efforts on integrating Netscape's Enterprise Server [3.5] because that's whatmost ISVs are sitting on top of today."

It's not just about internal cost savings. "Service classes for both users andapplications will give CIOs new tools with which to manage network resources and an[electronic site's] new pricing models," according to recent studies from ZonaResearch (Redwood City, Calif.) "Both types of networks will need to be able to sell'units of work' to different classes and prioritize and price such units appropriatelybased on user privilege, application priority and time of day."

Gold Medal Standards

HP isn't the only vendor working to ensure availability for commercial Web sites. Forthe most recent Olympic Games IBM developed software that balanced workloads across anumber of Web servers. At that time IBM concentrated on software that would help ensurethat a group of Web sites could act together to serve exceptional and sudden increases intraffic. By dividing the workload among many servers, IBM made their site reliablyavailable to Olympic fans.

IBM has improved on the technology, but unlike the separate focus on WebQoS by HP, IBMhas added-on to their WebSphere product. Called WebSphere Performance Pack, it's available(for now) on a wider set of platforms than HP's approach: IBM's AIX, Sun Solaris andWindows NT.

Eventually, Internet standards evolving around QoS and Differentiated Services willallow organizations to give preferential treatment to their most important customers; meetservice level agreements; establish and market classes of service; and eliminatebottlenecks due to overloading of key system resources.

By more effectively using current wide area network links and factoring a percentageincrease in Web-based business many businesses could find a nearly immediate cost savingsfrom implementing policy-based Quality of Service strategies. And it may be enough to getyour Internet customers to experience your site in real time.

--Mark McFadden (mcfadden@ is HP Professional'snew Consulting Editor.


HP's WebQoS software (now bundled with HP's new N-class servers) makes three major changes in the way you can view services on the Web:

The Creation Of Classes Of Users

Separating users into classes reflects the fact that, either inside or outside an organization, not all customers are equal. Providing a mechanism to identify different classes of users -- and then provide customized levels of service for those customers -- can mean the difference between satisfied customers and client contempt.

Classes of service, for instance, might divide the audience into tiers of customers: some getting high-quality, assured service, others receiving preferential service and the remainder getting best effort services. Another example is differentiating buyers vs. shoppers at an e-commerce site. Once a user has put an item into their shopping cart, they could be directed to URLs that provide enhanced services or better performance. According to Stan Schatt, Analyst at the Giga Information Group, "Using this scheme, the best customers can receive an even higher level of priority."

The Creation Of Differentiated Services

Something Mr. Murbarat would appreciate, is this approach to differentiate the services offered by a Web server rather than the treatment for individual users. For instance, users simply browsing back issues of the company newsletter would get a very basic level of service, users of an order entry system would get a higher level of service and executives drilling through ERP systems might get an even higher level of service.

The ability to change the performance characteristics of individual services makes it possible to have scarce system resources flow from one service to another as business needs change. A Web-based hotel checkout system could be given priority from 6 a.m. until noon and then the priority resources could be transferred to the hotel's registration systems.

Manage Peaks In Utilization

Peak Usage Management provides the ability to guarantee that existing activity on a Web server is not compromised and that new users are not admitted to a site before there are resources available to serve them.

Many Web sites know that they will experience spikes in utilization, but it's inefficient to over provision for those spikes. One example is the sudden stock slide in October 1997. As the market began to tumble downward online brokers started to have trouble keeping up with demand on their Web servers. Their customers cursed the network that was causing worse financial havoc every minute, but it was the Web servers, not the network, that were the source of the problem.

-- M.M.