A well-tuned Web machine
- By Brian Robinson
- Aug 12, 2002
The Web has turned out to be a wonderful means of delivering information and services, but it's becoming a bear to manage. The "old school" focus on keeping the network up and running is just part of the process. Maintaining the health and status of Web applications is the new worry.
And it will get worse. As Web applications become more distributed and complex under the emerging set of software tools called Web services, the traditional notion of unassailable borders of responsibility — the networking group has nothing to do with the database group, which has nothing to do with the storage group — will be overturned. All activities will influence one another, and keeping a Web-based infrastructure humming will become the ultimate balancing act.
Help may be on the way in the form of emerging performance management tools that aim to focus administrators on the outcome that matters most: end-user satisfaction.
The fact that traditional network management tools have not kept pace with the growing complexity of the Web environment has become a major headache for administrators.
For example, the old method of regularly "pinging" or polling a network using Internet Control Message Protocol packets to see whether specific devices and sites are up and running doesn't work in the new era of distributed Web applications, according to Miguel Rivera, project manager with the U.S. Customs Service's Network Engineering Team.
"You can't see what is happening with the end-to-end connectivity," he said, and that is vital because administrators must isolate the specific place where something has gone wrong in a large and complicated environment. Then they have to judge the effect of that problem on other parts of the infrastructure.
"Performance management in an n-tier environment is very difficult, simply because of the complexity introduced by all of those tiers," he said. An n-tier environment is one with multiple tiers.
The good news is that improvements in the technology of network devices, and the solutions used to manage them, are reducing the need for fault management. It's less likely these days that Web sites will actually go down.
The emphasis now is on improving the performance of those sites and the services delivered through them. Reducing the response times that frustrate Web users is becoming a priority.
And that's what performance management is about. The ultimate goal of user satisfaction — a Web site is always available and links are fast and accurate — translates to making sure the Web environment is available 24 hours a day and that the response time is always optimal.
TeaLeaf Technology Inc. is among the companies seeking to stake a claim to the new performance management territory. Its IntegriTea product monitors what the company calls the functional integrity of an application. Even when everything seems to be humming along nicely from the manager's point of view, the user may still be having a lousy online experience.
IntegriTea monitors user sessions and collects related transactional data from all available log files — the performance reports generated by the various components of the Web application, such as network equipment, databases and application servers. Then, in real time, the software summarizes system performance during many separate user sessions and presents the information to managers in a single view.
Performance problems could be caused by logical errors or by "other problems that are stopping the users from doing what they want to do," said Tim Knudsen, TeaLeaf's director of marketing. "Organizations usually don't have this user session information available when they are trying to solve problems just by looking at them from an internal management point of view."
Fortel Inc.'s SightLine takes a similar approach to collecting and aggregating real-time information on e-business transactions, including information injected into the process by an organization's business partners. It monitors what Fortel calls the "critical path" an e-business transaction takes across all of the application components in a Web environment and then applies a correlation engine to identify the relationships among those components.
If the performance of the system starts to move outside certain limits, SightLine can flag managers about possible negative trends, enabling them to correct the situation before users see a significant decline in service levels.
When things go wrong in a Web environment, according to Asa Lanum, Fortel's president and chief executive officer, managers typically try to pull together information on the various elements that make up a Web service and see how well they interact. But he said that's an often laborious, manual process that involves some guesswork, which his company's product is trying to remedy.
A 'Process' Solution
The trick for these and other tools, said David Moyer, a senior manager with KPMG Consulting Inc., is providing a solution for a process rather than for a point problem.
"The supply chain guy is looking at it from the perspective of the process, while on the other side, the IT guy is looking at it from the perspective of network connections, security, bandwidth, database access and so on," Moyer said.
It could be another 18 months to 24 months before most companies have their products at the level where they can provide the kinds of analysis that will accommodate those multiple perspectives, he added.
And it may take longer than that for the demand for those products to really show itself, according to Stephen Elliot, director of systems, applications and storage management software at the Hurwitz Group, a market research company. The general demand now is for basic, tactical tools that will allow managers to wring more value out of the monitoring and management tools they already have, he said.
"Some of these newer tools are very innovative, and there's certainly a lot of investigation of them going on," he said. "But there's a lot of confusion among users about what is truly needed to improve the performance of applications and the way they fit into service-level requirements, what impact they will have on databases and so on."
Robinson is a freelance journalist based in Portland, Ore. He can be reached at firstname.lastname@example.org.
Brian Robinson is a freelance writer based in Portland, Ore.