Implementing Service Level Management

by Jean-Pierre Garbani

How Businesses Use Technology: Pushing the Envelope

The IT infrastructure and the corresponding organizational structure exist solely, in any enterprise, for the purpose of supporting the business processes. Over the years, the advance in technology has allowed IT to "push the envelope" of how businesses use technology:

• The number of business processes that could benefit from IT, from the simple, objective of administrative support has grown to include to more subjective areas such as marketing and sales.

• The areas of a business process that could benefit from IT support have expanded beyond data processing to provide decision support information, e-business, and so on.

Studies have shown [Schneiderman 84, Thadhari 81, Guynes 88] that the key to productivity in interactive applications is a rapid response time. As long as the user and the system keep a compatible pace -- that is they don't have to wait on each other -- productivity increases, cost of work drops, and quality improves.

In this environment, the IT management role is to provide not only the functions required by businesses, but to provide them with a satisfactory level of performances. The cost of providing these functions and keeping the user satisfied has to be minimal, or at least competitive with outsourcing providers.

The problem of reaching these objectives has multiple aspects:

• The non-linear growth of user demand
  As long as user functions are confined within a limited number of processes, growth is a linear function with roots that are in the business growth and the corresponding increase in, for example, personnel. When the paradigm shifts from data to information, the growth becomes non-linear since it shifts from a process-based demand to an individual-based demand. The example of "Web Surfing" shows how unpredictable this could be.

• The lack of suitable management tools
  For a number of years, infrastructure management vendors have provided a number of tools to capture and report component behavior, on the basis that "healthy" components must produce good performances when grouped together. The problem has proved to be more complex, however, and requires a more sophisticated approach.

The simplest, most common form has been, in most IT organizations, the creation of a "Fire Brigade." In firefighing mode, through help desk, trouble tickets, and the like, the IT organization basically waits for the user to complain before identifying problems and correcting them. Network and system management tools provide real time information about infrastructure components to the control center function. This leads to several potential problems:

• Users are experiencing a roller coaster ride in terms of performances, the net result being a lack of confidence in the infrastructure and the IT organization.
  

• The costs of constantly "fixing" problems under time pressure is higher than it should be in a planned mode.

• The IT management does not have any visibility of the long-term deficiencies of the installed infrastructure and lacks the basic information needed to plan for its evolution.

The correct way to go about all this is to bring the IT infrastructure into a really controlled and managed environment. To achieve this, the IT management will have to control two key elements:

• user expectations in terms of performance, and, of course, costs. This means that, instead of blaming each other for the current results, both parties (users and providers) agree on a certain level of performance and costs.

• performance of the IT infrastructure has to conform to the agreed upon level. This forces IT to be managed in a "predictive" rather than a "reactive" mode.

Improving Infrastructure Management: Shifting Gears

The first step in improving the situation will be to shift from an essentially subjective world to an objective one. By involving the user in the cost/performance trade offs necessary for a certain level of performance, the IT organization can bring user expectations to an objective and realistic level that is accepted by both sides.

The Service Level Agreement (SLA) is the instrument resulting from this negotiation. It provides a yardstick by which actual measures of performances can be judged, costs compared, and improvements negotiated.

The key to appropriate and sustained performances, and the basis of Service Level Management, clearly resides in IT management's ability to:

• create, by capturing key metrics, a performance scorecard at component level

• aggregate it into a user performance scorecard in line with the SLA

• manage the infrastructure capacity in such a way that the balance is maintained between cost and performances

The success of this enterprise relies on two capabilities:

• The ability to create component models, which, from metrics captured at the IT infrastructure level yield an exact performance parameter such as service or response time.

• The ability to aggregate thousands of these parameters into user scorecards.

The following figure illustrates how service level management can be implemented:

© Giga Information Group, Inc.

The first three layers of the model shown above are the data capture layers. All components of the infrastructure are represented, from hardware or service components such as server and network, to storage components (databases) applications. Different web services, which could be a platform for browser-based applications (a J2EE or .NET platform), or external services used in the performance of an application are also included in the data capture layers.

The second layer shows the parameters to be captured. Availability and performance, traffic and response times are pretty straightforward. Increasingly, parameters such as "application accuracy" and "security" become relevant. The content of a Web page, for example, becomes as important as the capacity to deliver it on time. Reacting to security breaches in an infrastructure that is open to the Internet is also a key point.

The data captured is reported in real-time in a Network Operation Center (NOC), where it is used in real-time to detect potential problems (alerts, alarms) and to identify their root cause to determine a real-time corrective action. Load balancing, alternative sites, and now utility computing are the types of corrective action available in modern computing infrastructures.

The Pivitol Layer

The pivotal layer in this infrastructure management model is the service level management layer. The data capture layers are traditionally component or device-oriented. Also, the functions provided are the traditional functions offered by the network and system management products, either through a framework, a suite of products, or a series of point solutions.

At the SLM level, however, the report on infrastructure performance becomes user-oriented instead of device oriented. This is where the different parameters collected are aggregated into end user "scorecards."

This demands an aggregation model. Since the link between the business process and the infrastructure is the application, reporting service levels as perceived by the end users requires that models of how the infrastructure supports the application be built, as illustrated in the following figure:

© Giga Information Group, Inc.

Most products available at this level propose a manual aggregation of the components. However, there are currently a number of companies working in infrastructure administration and configuration management that are on the verge of providing an automated way of "discovering" the components of an application and the dependencies between them. This will be a giant step forward in the implementation of SLM.

The upper layers of the infrastructure management "stack" are all capitalizing on the ability to build these models. Performance management is the capability to use the model to determine performance bottlenecks in real-time. Capacity planning is the ability to forecast the evolution of infrastructure usage or the impact of new applications and to adjust the infrastructure capacity accordingly.

Finally, the business process performance management is an aggregate of all data received from the application models in order to create a view of how IT is serving a business process in terms of efficiency, effectiveness, and costs.

Implementing service level management consists essentially in creating management processes within the IT organization that are capable of using the data provided by a number of tools and to convert that data into information. When an infrastructure has been instrumented in such a fashion that it can be controlled in real time from a NOC, a cross-divisional process has to be created within IT Operations to aggregate data in a way that is meaningful to the end user. Once this is done, the process can be capitalized upon to open the door to a complete control of IT performances and costs.

Jean-Pierre Garbani is a Research Director for Giga Information Group, Inc. His area of focus is on the performance and capacity management of the IT and Web infrastructure. Jean-Pierre has more than 30 years of experience in developing and integrating a number of technologies including real-time systems and very large distributed infrastructures for the government and banking industry in France, Scandinavia and the US. Prior to joining Giga, Jean-Pierre was director of capacity planning at John Hancock Financial Services, in Boston. He has also held a senior consulting position with Onsett International, as well as different system development, marketing and sales positions at Bull Information Systems, in the US and France.

home