Managing Applications from the User's Side
To better understand the challenge of application management, eavesdrop on the calls the helpdesk often gets.
Caller: "I'm having problems with SAP. It's real slow today."
Helpdesk: "Have you changed anything with your system? Have you added any new software or anything?"
Caller: "Nope. Nothing. I'm doing what I always do but it's real slow. I mean really slow."
It is calls like these that drive technology managers nuts. "Users call up and complain that response from some application is slow," complains William Chin, director of global technology at Canadian Imperial Bank of Commerce (www.cibc.com). "We use RMON, SNMP. We look at bandwidth, but we can't find any problem. So, maybe we try to ping the application. We find it takes 40 msec. That's not a problem. But the user says it's taking two minutes."
The problem is that for all the investment organizations have made in network and system management during the past decade, "We still don't have a way to see things from the user's point of view," Chin continues. The support staff can look at individual pieces -- the network, the servers, the database, any number of components --but still not experience the application end-to-end as the user does.
Application management -- different from system or network management -- promises to give the technical support staff the ability to monitor the application end-to-end, track key performance metrics and take action when necessary. It promises a way to monitor, measure and manage the performance of the application itself, not the network links, routers, hubs, servers or other system and network components.
In comparison, network and system management focuses on the links, network devices and servers. It tells a manager about dropped packets and CPU utilization and about unresponsive devices and memory constraints. But it says nothing about why the e-mail is taking so long.
"Users don't care about pinging," says Patrick Dryden, senior industry analyst at Giga Information Group Inc. (www.gigaweb.com). "They don't care about packets or CPU utilization. They care about what they see -- the application, not the plumbing." With the introduction of distributed, mission-critical production applications, such as ERP solutions, IT groups are under the gun to manage the applications for top performance. "These applications have much higher perceived value, and once they are up and running they become critical to the business," Dryden adds.
Hand in hand with widespread adoption of larger packaged business solutions is the trend toward service level agreements (SLA). An SLA is a contract that specifies levels of performance and availability for key applications that IT commits to provide. Application management is a key mechanism by which the SLA can be monitored and enforced.
Application management, as a distinct market segment, is in its infancy. Vendors and analysts estimate the market to be in the $100 million range this year. Numerous small players such as First Sense Inc., Ganymede Software Inc., NetIQ Corp., INSoft, Inverse Network Technology Inc., NextPoint Networks, Response Networks, and Jyra Research Inc. populate the market. Big vendors, such as Computer Associates Int'l Inc., IBM Corp.'s Tivoli division, Hewlett-Packard Co. and BMC Software Inc., also sell products that could be considered application management tools, which would push the market estimates higher.
As Windows NT takes on more roles in the enterprise, including hosting critical databases and core packaged business solutions, the need for application management increases. Windows NT's ability to monitor and record almost every cycle of the CPU in voluminous logs, provides a foundation on which to build effective application management. As a result, application management tool vendors are rushing to introduce tools for the Windows NT platform.
Don't make the mistake of thinking that only complex applications, such as running SAP, will benefit from application management. Running Microsoft Exchange across a handful of servers can quickly exhaust an application management team. "Exchange churns out thousands of events which are put in the log. Microsoft recommends looking at event logs each day --that's an impossible task," says John Matthew, messaging technical support lead of global messaging applications at Universal Studios (www.mca.com). Sorting through thousands of events on each server each day to spot problems and potential problems is harder than finding the proverbial needle in the haystack.
Charles Schwab & Co. Inc. (www.schwab.com) faced the same problem. With over 600 servers, the company needed something to manage its critical NT applications, especially Microsoft Exchange. The company turned to NetIQ (www.netiq.com). "We can't do point management of every server. It would simply take too much labor," explains Bharat Patel, managing director of enterprise management. NetIQ, however, sorts through the NT events and alerts managers to problems. "A NetIQ shows us only what we need to fix," he adds.
With few exceptions, most application management tools don't fix anything. "For the most part, these are monitoring tools. They let you take a more holistic view," Giga's Dryden notes. The guiding principal is that once you know how an application is behaving, you will understand what actions need to be taken, whether it is reordering queues or balancing the load between multiple servers.
"The goal is to automate actions in response to information," adds John McConnell, president of McConnell Consulting (www.mcconnell.com). Now, most tools simply report things. BMC's (www.bmc.com) tools, he notes, are an exception because they can adjust to a monitored application dynamically. Other tools provide the ability to automate responses to specified events, but this usually demands scripting on the part of the manager.
Union Gas Ltd. (www.uniongas.com), for instance, uses BMC's Patrol to monitor activity around Microsoft Exchange. Patrol sorts through the information and alerts the managers of problems, such as when queues grow too large, says Scott Jacklin, senior technology specialist at Union Gas. Once the Exchange managers receive an alert, they initiate corrective action. Patrol provides a few automated scripts for the most common alerts.
Application management has attracted a host of vendors. "Starting late last year and exploding into the beginning of this calendar year, one vendor after another has proclaimed its interest in providing this [end-to-end application] view," reports Hurwitz Group (www.hurwitz.com). The result, according to Hurwitz, is a confusing array of solutions offering "a plethora of views displaying the response times that the end user experiences on the server, on the network and on various constituent devices."
To help managers sort through the many application management tools, Hurwitz Group organizes the tools into three general areas: early warning trouble detectors, problem analyzers and managers.
Early warning trouble detectors give advanced warning of performance degradation by monitoring specified network, server and application metrics, and by triggering alerts as they approach previously defined thresholds.
Problem analyzers identify what went wrong and suggest remedies. In some cases, the tool may actually to the fixing.
Managers provide information integration and reporting on a device, link or the state of an application.
Almost every application management tool monitors and collects information through the use of an agent. Some tools, notes Giga's Dryden, are network oriented. They place probes at points in the network to try to determine the performance of an application by analyzing the bits that pass by. Other tools monitor the action from the user workstation.
The end user agent can be either an actor or a watcher, Dryden explains. Watchers follow every keystroke a user makes. Actors are robots that run a sample transaction and monitor and measure the response. However the tool works, organizations have to be selective about how much and how frequently they monitor, or they risk impacting application performance and overloading the network. There is also the danger that the tools may flood already harried managers with additional volumes of data.
The Kansas Department of Administration, for example, uses Compuware's EcoScope to manage application performance, but managers are careful in how they use it. "We use it selectively. When we get a complaint, we will put in an EcoProbe and find out what is not responding," says Swanee Lawrence, programmer analyst at the state administration. The probe tells Lawrence if the problem is with a Novell server or an Oracle application. It will even dig down to the SQL verb level to uncover a problem.
The cost of application management can be lofty if the buyer isn't careful. "Companies have to look carefully at the pricing," Dryden advises. Some vendors give away agents for free but charge for servers. This encourages companies to place the free agents on every desktop and device. The result, however, may be the need to buy larger or multiple servers, which increases the cost.
Other vendors charge for agents. In that case, companies will be more selective in deploying agents if they must pay for each one, which cuts down on their management ability. Then "there are a lot of hidden costs to application management -- the care and feeding issue," McConnell adds. Like system and network management, application management does not come cheaply.
Two initiatives are under way to simplify the application management process. The first, called the Application Performance Measurement forum, is a consortium of small application management vendors. Their purpose is to define unified terminology to ensure terms, such as application management, end-to-end response or service-level management, are used consistently by vendors. It is about standardizing a vocabulary, and if everyone complies, it should make selecting and evaluating tools easier.
The second initiative is a joint effort between J.P. Morgan & Co. Inc. (www.jpmorgan.com) and Computer Associates. This effort is attempting to define a standard application management programming interface, called the Application Management Interface (AMI), as a public standard. "The AMI will be a way to get data out of the application to business operations managers," explains John Elvers, vice president and project manager for the AMI effort at J.P. Morgan. The AMI, intended to complement the application response management (ARM) initiative spearheaded by Tivoli and HP, will give software developers a way to build applications that can report behavior to management tools and respond to instructions received from management tools.
Application management represents the next stage in the evolution of distributed computing. It extends the discipline of network and system management to the application. Windows NT, with its penchant for logging everything, provides a start for application management tools. But the real challenge is to get beyond the NT server and experience the application the way users do. For this, an application management tool will be a necessity.