Given the transition from big iron to Linux, traditional systems management methodologies are no longer universally applicable. In traditional big-iron data centers, management is based on the centralized deployment and control of a homogeneous environment. Linux management requires a distributed deployment and management model. Traditional management solutions attempt to address this issue by relying on an old approach, based on the network installation of an image, often referred to as the “golden image.” Golden images can be installed on any node. Any subsequent changes to the system are based on writing and running management scripts, which can be time- and labor- intensive. Provisioning and patching, change tracking, auditing and compliance, and backup and recovery become arduous tasks, requiring explicit administrative attention. Overall, image-based management (like the golden image approach) requires too much time, making resource management difficult, migration and repurposing of servers tedious, and recovery from failure and energy conservation almost impossible.

Modern Linux data centers are replacing image-based management with a three-pronged approach, based on state-based management, data center automation, and virtualization.

State-Based Management – Moving Beyond the Golden Image
State-based management, as its name suggests, looks at the state of a system. The software stack is seen as a static state, comprising the operating system, application packages, and configuration data and scripts, while the execution data is seen as part of a dynamic state. Under a state-based management approach, the state lives in a repository in shared storage. The repositories are shared by all servers in the data center.

Provisioning and Patching
Using state-based management methods, provisioning involves binding the static state, or the software stack, to the hardware, creating references to packages in the repository so that provisioning changes can be tracked and monitored. When patches are issued and installed, system administrators simply add updated packages to the repository and update each server’s state. While this process may be initiated manually, pushing patches out to the servers in the data center is no longer as arduous as it once was with image-based techniques.

Change Management
State-based techniques also improve change tracking, because every change in the state is saved. Since each version of the system’s state is saved, a server in the data center can be easily restored to any previous state. Because the state-based approach tracks the incremental changes of each individual server, it has the speed and portability to handle the demands of hundreds, even thousands, of servers.

System Recovery
In the case of a server failure, the recovery process is efficient. Users can migrate saved server states by unbinding, or breaking the association between the server and the hardware, then rebinding to new hardware.

In essence, state-based management can ease various challenges in content management, change management, and resource management. This management approach is excellent for provisioning, change tracking, and auditing and compliance, while operations such as patching, back up and recovery, recovery from failure, migration and re-purposing are all significantly simplified. And, unlike image-based management, the state-based approach requires no disks, which can vastly improve reliability and energy conservation in a data center.

While a shift to state-based management brings the system administrator closer to the goal of good data center management, some of the tasks, while no longer as painful as they once were, must still be initiated manually. For data centers composed of larger server farms, this necessary human touch can add significantly to the workload of a system administrator, who must detect conditions and decide on the appropriate course of action, with the management system used only to implement the changes.

Policy-Based Automation
Automation can reduce the workload of an IT professional in several ways, but first let’s focus on the key elements of automation. Automated lifecycle management should include both a monitoring system to detect events in the data center’s environment as well as a policy engine. A rules-based policy engine can take appropriate actions based on the monitored events. Any changes to the servers and infrastructure in the data center are picked up by the monitoring system, and the pre-configured policies provide for flexible, responsive, automatic lifecycle management.

Monitoring systems must be able to detect and report changes in hardware, software, network and storage infrastructure, and the management system itself. It also has to monitor the physical aspects of the environment, such as power levels and cooling. Each change should be reported as an event, which can then be interpreted by the policy engine.

The policy engine should be composed of business rules that are expressed as policies that dictate the course of action for a range of events. The policy engine listens for events signaled by the monitoring system and, having been alerted to an event, generate corrective actions based on the business rules. The engine serves to automate routine lifecycle management functions, reducing the degree to which system administrators must make decisions about how to respond to every single event in the data center.

Automation can be implemented so as to fully automate patching, change tracking, auditing and compliance, migration and re-purposing, while partially automating provisioning, backup and recovery, and recovery from failure. Combining automation with state-based management significantly reduces the complexity of essential management tasks, as well as the need for the immediate attention of an IT professional for routine operations and events.