Policy-Based Computing

The world of computing is about to change. As more and more devices become connected, the way the systems need to be managed must also change. Much of the focus in device technology over the next decade will focus on increasing the intelligence of the system, whether the system is a cell phone using Bluetooth to connect to a laptop or car or a new system like the data center enabled for the N1 Grid.

The N1 Grid software products concentrate on automating many of the manual tasks performed by administrators, while providing virtualization and automated provisioning of resources. Today, most of these tasks are initiated by a human, based on some specific set of policies. These policies can be written down, can be in configuration files, or perhaps are only kept in a person's head. System vendors, including Sun, are rapidly moving towards enabling these policies to be specified in business terms so that the holistic system can manage itself to these policies.

The keys to policy-based systems are three-fold:

1. The ability to regulate the system (usually known as the manager)

2. Ability for the system to know itself and to monitor its condition (usually known as sensors or telemetry)

3. Ability for the system to know what it should be doing, based on activities or state (that is, the policy)

The automobile is a great example of policy-based systems. Cars used to be controlled solely by a human, with all the decision making left up to the driver, in real time. The modern car is changing all of that. Consider some of these changes over the last decade:

• Engines now have several computers that continually adjust the fuel mixture, cooling, and electrical system.

• Antilock brakes are almost standard equipment, continually sensing the road and driver, and adjusting braking power as needed.

• General Motor's On-Star service has been highly successful in enhancing the safety and security of the car. Operators can remotely diagnose a car, while it is running, and determine if and when a repair is necessary. If the air bags deploy, the system can automatically contact emergency services. These are all great examples of enabling the feedback loop. The system has been closed by automation, policy, and optimization.

Consider the bookstore discussed in Part 1. The bookstore service has some specified policy around it. It has an availability goal, which is translated by the N1 Grid software control plane into various technical specifications on how the service is deployed and managed. Should a system become unavailable and threaten the availability of the service, the N1 Grid software automatically adds another instance, and generally, the user is not aware of anything happening.

Policies can also describe security, such as which systems and what storage users can access, how much disk space an application can get, and how to respond when it needs more. Policy is everywhere in the data center, which highlights the difficulty of this area. The first phases of the N1 Grid vision concentrate on the control aspects of the system (the manager part of the three-part policy system). Automation around system optimization and descriptive and prescriptive policy are not far off. In fact, the N1 Grid PS has rudimentary capabilities in policy. You can already describe a minimum set of servers within a server group, and if a system becomes unavailable, another one will be provisioned from the pool.

Feedback Loop

The key to creating the policy-based data center is in enabling the feedback loop. There really is nothing new here; computer scientists have been dealing with autonomous systems for decades. In fact, many aspects of policy-based systems start to look much like biological systems and biologically inspired computing.

Managers

Policy-based systems need managers. The N1 Grid SPS and N1 Grid PS are examples of products that provide agents to manage the system. If additional system capacity is needed, the administrator can easily instruct the N1 Grid software to add the additional resources.

Sensors and Observability

Policy-based systems need sensors and data. For the human or the system to make the correct decisions, they must have data. Several tools can be used to outfit the data center to enhance the telemetry. These tools are critical for policy-based systems and the ultimate N1 Grid vision.

Policy

Policy-based systems need policy a set of constraints and rules in which to operate. These policies need to cover actions and activities, such as when something changes within or external to the managed system. This means policies must describe how to handle changes in state of itself or another nearby system, whether the system can do something, and how it will do it. FIGURE 11-4 shows the policy-based computing feedback loop.

Figure 11-4. Policy-Based Computing and the Feedback Loop

A simple example of policy is the profiles within a cell phone. Users can tell their cell phone how to react when someone within a certain group calls, and they can set up various user profiles based on time or other connected devices. When the cell phone changes to a quiet state, a different set of policies controls its behavior.

Policy-Based Computing and the N1 Grid Software

Before you can increase the observability system to capture system changes and to train itself to do something, a system needs to be developed to document the policy. Policy is distributed, and each subsystem might be responsible for some of these activities.

Capturing and Describing Policy

Services-based management, including the N1 Grid software, and feature-rich monitoring and control systems are required to enable policy-based computing. Many of these capabilities are in the N1 Grid software and will be enhanced to provision data center optimization capabilities. Today, IT staff need to first automate many of the common activities, focus on higher-level policy, and work with companies such as Sun to provide the correct implementation of policy-based standards.

Preparing for Policy in the Data Center

As with any new technology, grid computing and utility computing are both emerging areas, much like the N1 Grid technology. However, they provide excellent platforms to produce real results and fit well with the N1 Grid software management model. Companies today are managing grid and utility computing systems using the N1 Grid software. In an online article titled, "Tech Wave 1: Utility Computing," Business Week considers utility computing to be the next revolution after the mainframe and the Internet, and as standards progress, utility computing and grid computing could appear to be much alike.