9.1.1 Dynamic operations
Dynamic operations consists of autonomic managers that unify the infrastructure to maximize utilization through defined business goals. These autonomic managers monitor performance metrics, analyze the monitored data, offer a plan for running various actions, and can start these actions in response to the flow of work.
Extended Deployment offers the following autonomic managers as part of the functions of dynamic operations:
- Autonomic request flow manager
Controls the order of requests into the application server tier and the rate of request flows. Using classification and the defined service goals, the autonomic request flow manager decides how and when to dispatch HTTP requests to the next tier.
- Dynamic workload manager (DWLM)
Performs load balancing across available application servers. In particular, for a given request flow, DWLM balances requests across the available nodes to regulate response times. DWLM dynamically updates the application status as the application placement controller modifies the infrastructure for the application that is running.
- Application placement controller
Creates and removes application instances to manage HTTP requests. The application placement controller can dynamically address periods of intense workflow that would otherwise require the manual intervention of a system administrator.
- Health management
Uses health policies to identify the criteria that require action. When the criteria are met, action is taken to ensure that the environment remains healthy.
These autonomic managers and the on-demand router (ODR) are the primary functional parts of dynamic operations. An ODR can be defined and started before any service policies are defined, but operational policies can be defined before the appearance of the work to which they apply. However, if policies are not defined, the early work is handled by the default policies.
When work enters the ODR, an optimization effort achieves a balance of performance results. As the work flows, the dynamic workload manager balances the load. As work variations change and the balance of work in the nodes is upset, the application placement controller, autonomic request flow manager, and dynamic workload manager rebalance applications that are running to ensure efficient workflows. The combination of these autonomic managers provides a seamless, end-to-end dynamic runtime ability.
The autonomic controllers within WebSphere XD make use of a number of estimates, but the following two estimates are the most relevant to the integration of DPLARs:
- The quantity of resources required by each request in a given service class is estimated by a work profiler, which updates its estimates as requests are processed.
- The WebSphere node agent that is running on each node estimates the quantity of resources including memory capacity and CPU bandwidth that are available at that node. The node agent publishes information about the available resources by following these procedures:
- Entering that information into a distributed data structure that is accessible to the WebSphere XD autonomic controllers
- Sending updates of that information to WebSphere XD components that request them.
With WebSphere XD, when the node agent is running in an AIX DLPAR, it periodically monitors the CPU and memory resources that are available to the partition. When resource availability changes, the node agent updates the WebSphere internal data structures and actively notifies any components, including the application controller, of the new resource levels.
Because the autonomic managers dynamically adjust their calculations as those values change, the routing and application placement decisions that WebSphere XD makes in the clusters and the applications that it manages quickly reflect the new allocation of resources between partitions.