|
Agent-based software resides on a remote IT object (such as a server) that collects data based on policy set locally or by the management server. These agents observe the state (or collect and aggregate performance data that causes an event to be issued) of IT objects and communicate any detected state changes to the management server, where management policy is set. Typically, the agents communicate with the management server only if policy breaches are detected (for example, when a performance or storage capacity threshold is being exceeded) or at pre-scheduled times to upload historical data. Thus, the impact on the management server is a lot lower than on an agentless management server (an agentless management server typically does all the polling and data filtering) and so is the impact on the network (no constant polling for data). In addition, an agent-based product can provide a level of IT management autonomy, because the agent can cache collected data, even if the management server or its connection is lost, and can then send the data once communication with the management server is restored. Thus, the agent can take action regardless of the management server's communication availability.
Agentless monitoring can be approached in two ways. In one way, data is made available via remote application programming interfaces (APIs) — for example, Simple Network Management Protocol (SNMP) Management Information Base (MIB) for network devices, Windows Management Instrumentation (WMI) on Windows devices — thereby negating the need to deploy an agent to a server. The other way to approach agentless monitoring is by analyzing packet flow across the corporate network infrastructure with a sensing system (either hardware or software) located at a strategic point in the infrastructure's topology and capturing information about response time, availability and other flow-oriented functions as traffic goes by the sensing point. This approach is more in line with monitoring end-user response times and with root-cause analysis, and typically is not used to monitor hardware and software.
Although the industry historically has opted for placing agents on the servers they are intended to monitor, Gartner increasingly is finding that many clients are selecting agentless approaches, which are easier to implement and administer for monitoring. Much of the data that clients seek is available via remote APIs. Typically, an agentless approach doesn't consume as many resources on a server being monitored, although remote polling has some impact on the server and network.
The short answer to this question is no. For mission-critical applications, end users still will want to deploy agents that can continue to perform their activities, even if communications are disrupted with the central console. IT organizations must be aware that the builders of the underlying infrastructure (BEA Systems, Cisco, IBM, Microsoft, Oracle, SAP and others) are continuing to add management functionality to their platforms that is available via APIs. As more instrumentation is embedded within the infrastructure, we will find less need for agents on remote servers — at least for the more-mature platforms. With the introduction of new technologies (such as Web services), we typically observe that it takes the vendor several years to embed robust manageability. For these platforms, agent-centric approaches from a third-party vendor will be more viable.
Traditionally, enterprise-strength agent-based monitoring products have two-tier (management server and agents) or three-tier architectures (management server, midtier managers and agents).
Midtier management products can provide a number of capabilities, including the storing and forwarding of event data and, in some cases, the ability to set policy on the agents. Even though the midtier manager can provide a level of scalability through scale-out architecture and resilience (for example, midtier servers providing failover capabilities), the main value of the midlevel manager is to enable the agent-based product to perform and scale.
Agentless management tools typically are single-tier, (similar to most SNMP network-monitoring tools) where the management servers are located strategically around the IT infrastructure, pulling information from the APIs (network, servers, applications and database). Although agentless tools can be deployed quickly, the need remains to set up policy that ensures that the central management server does not become inundated with data. In addition, if resilience is important, then effort must be invested in making sure there are no single points of failure (if not available through the tool itself, then utilizing clustering or other high-availability technology).
|
