App-titude adjustment

Network pros sink their teeth into managing applications.

As recently as two years ago, the typical network organization had little if any responsibility for managing application performance. But that situation has changed, and now the majority of network organizations are actively engaged in application management .

“Because it is common to blame the network for poor application performance, networking organizations have no choice but to be actively involved in applications management,” says Terry Dymek, senior director of EMC’s  internal IT organization.

To quantify the state of the art in app management, Ashton, Metzler & Associates surveyed more than 100 network professionals. We found that network organizations have made great progress relative to applications management, but there is still a long way to go.

It’s all about the apps

According to the survey respondents, a primary reason that applications management has become so important is that the majority of business managers see the value of IT as coming primarily from applications, rather than the supporting infrastructure.

Karl Wagner, director of global network and telecommunications for PricewaterhouseCoopers’ internal IT organization, puts it this way: “The network is regarded as the pipes and sewers of a city, while applications are shiny, glass, architecturally pleasing buildings that everyone sees above ground.”

Another major survey find is that IT professionals regard applications management as the most difficult component of network and systems management. To respond to the importance and the difficulty of applications management, successful enterprise IT organizations use two related approaches. One is to model and profile applications to quantify applications performance. The second is to implement ongoing applications monitoring.

Applications modeling

This technique lets IT organizations simulate the performance of an application before deployment. It would be nice if there was a generic model that provides meaningful insight into the performance of an enterprise application such as SAP. However, this goal is difficult to achieve because of the level of application customization that typically occurs.

In particular, most companies make changes to standard applications before they are deployed. In many cases, even a minor change can result in the modified system behaving significantly different than the standard application. Wagner points out that each SAP implementation is unique and hence performs differently.

Application profiling

Dymek says the generic information he gets about application performance from major software providers such as Oracle and Microsoft  doesn’t help much when it comes to his specific network. So he has instituted a program by which his organization works with the application developers to profile application performance at various stages in the application development life cycle. According to Dymek, application profiling lets him set appropriate expectations as to how a given application will perform on a worldwide basis. In some cases, the insight gained by profiling an application also has caused the application developers to change the software.

Ongoing applications management

Our data (see graphic, right) clearly indicates that network organizations have come a long way in terms of application management. For example, while 22% of organizations still throw capacity at performance issues, that technique is not as widespread as it used to be. It’s a far less popular approach than using tools to manage and monitor applications, which 48% of respondents said was their primary technique.

One way that companies avoid throwing bandwidth at performance issues is to implement quality of service (QoS ). Dymek says his organization has implemented QoS throughout the Asia Pacific region to reduce the cost of expensive WAN links. QoS lets Dymek give applications from companies such as Oracle and Clarify the priority they need to perform well, while introducing only modest degradation in the performance of less critical types of traffic such as FTP and e-mail.

Dymek also says his organization has established goals for network latency and that they continually monitor the network to determine whether or not they are meeting their goals. However, Dymek pointed out that the monitoring they do is not application-specific in part because he has not been able to find a monitoring tool that can manage application performance, particularly of n-tier applications that use Web services.

“The network management suppliers are just now catching up with Cisco’s ability to provide QoS and provide corresponding queue-level performance measurements. Before QoS, standards management methods included SNMP polling and ping measurements. With QoS, these SNMP  polling latency and ping measurements cease to provide meaningful measurements. Network management suppliers must instead rely on queue measurements from routers,” Wagner says.

Our data also indicates that network organizations still have a long way to go. For example, while most organizations have tools and processes to manage and monitor applications on an ongoing basis, only 17% use these tools and processes in a proactive manner.

Pulling it together

Organizations that are serious about managing one or more applications should start by reviewing any models that have been developed that can provide insight into the performance of the application. Application profiling can provide details about how their implementation of an application is likely to perform. While techniques such as QoS can ensure network performance, the biggest challenge organizations will face is to implement tools and processes that directly enable the management of complex applications.

Metzler is vice president of Ashton, Metzler & Associates. He can be reached at jim@ashtonmetzler.com.