A sensible mixture of agent-based and agentless approaches is best, with the majority of the environment supported by agentless.

An agent is software instrumentation that presents information about the device, system or software program in which the agent resides.
An agent may be built in (e.g., SNMP MIB, Microsoft WMI) or "bolted on," as is the case with most of the Unix server agents.

When the agent is built in, the term agentless is used to describe the management of that element. This is a bit of a misnomer, however, since every management task requires some form of intelligence. Hence, everything that is managed has some form of agent. The only difference is whether it is already embedded or is an additional software component that needs to be installed.

A sensible mixture of agent-based and agentless approaches is best, with the majority of the environment supported by agentless. Agentless systems are simpler to deploy and maintain, but the most critical elements of an environment may justify carefully placed agents.

The growing complexity of IT environments and business services is fueling the drift away from agents and toward agentless systems. Just a few of these forces are:

Virtualized environments. Large server operations are increasingly virtualized, which means an agent no longer can work by just sitting on a physical server. In a virtualized and decoupled environment, systems management is no longer linked to specific physical servers. Given increasingly complex applications like Web services, the compute environment has quickly grown large and complex.

A dramatic shift toward distributed systems. Moving away from client-server applications to distributed n-tiered applications means adding more hardware and software all the time–with no end in sight. Deploying and maintaining agents on all of these separate devices makes management overhead demanding. With new devices coming online continually, agent installation and upgrades are never-ending, and different agents on the same device can wreak havoc with one another.

Integration and communication issues. Agents must co-exist with one another, a tricky operation at best when multiple agents from multiple vendors are running on a single target. The situation is even more complicated because agents from different system-management vendors will have different upgrade cycles, and a newly released agent upgrade will not necessarily communicate with another company’s agent.

Relationship mapping. An emerging trend in infrastructure management is to focus on service-level agreements between business application users and the underlying IT infrastructure. This trend is creating demands for management technology to address fault, performance, configuration and control issues across hardware domains, rather than within storage, systems and network domains. Today’s management agents, however, are hard-pressed to meet demands such as defining the relationship between a device and an application, or between two or more applications.

SOA destroys the status quo of management. The emergence and growth of service-oriented architecture (SOA) brings unprecedented flexibility, but also unprecedented complexity. Traditional agent-based management has difficulty coping with the relative quiescence of current applications and services. SOA shrinks the time scales for changes to relationships and configuration settings. Traditional management will collapse under these taxing conditions.

This last item highlights how the "flattening" of IT is being driven by the dynamic capabilities enabled by SOA and how management technologies should adapt to this new world. Future business processes, and their executions, are being coded in new SOA standards. Internal applications integrate with external partners, providing simplicity that has not been possible before.

Management systems should be integral to this trend. Historically an afterthought, management is becoming a higher priority relative to the technologies being managed. SOA and virtual systems cannot operate without the intelligence and controls provided by management systems. Rather than a cumbersome bolted-on attachment, management is becoming an internal part of these sophisticated technologies themselves.

Management intelligence embedded within a system is inherently agentless. The inclusion of management within the new technologies accelerates the trend away from agent-based management. This allows the flexibility and control that will be needed. Bolted-on technologies will act as an impediment to this flexibility and control.

Emerging standards are already bearing fruit (OASIS has published the Web services distributed-management specification–WSDM), although this transition is a multiyear initiative. The development of service-modeling language (SML) promises even more capability. Such semantic modeling efforts will become the core of management systems of the future and further marginalize the concerns about agents. Agents will be built into everything.

As WSDM, SML and other advanced methods come into widespread use, management will become truly real time. In the past, monitoring has been hindered because it relies on periodic polling.

Limitations in agents (whether add-on or embedded) require management servers to poll the instrumentation for most functions. These polling intervals are typically 15 to 30 minutes. Shrinking the polling interval to seconds rather than minutes gets closer to real-time, but such a move is considered impractical, as resource overhead increases exponentially as ever-smaller intervals are approached.

The right mechanism for real-time management is to place more management intelligence in the instrumentation and to have that intelligence notify the management server asynchronously. The resource demands of this model are more favorable and allow for the capture of conditions as they occur instead of waiting for the next poll. The embedded instrumentation that is now emerging in SOA applications and services (as well as infrastructure) is designed to use this notification model.

Glenn O’Donnell is senior product manager, resource management group, EMC, Hopkinton, Mass.

For more information:
www.rsleads.com/706cn-256