Wednesday, September 23, 2009

Oracle 11g Release 2 Clusterware New Feature: Policy-based Cluster and capacity management

I was reading through the Oracle 11g Release 2 Real Applications Cluster New Features List and stumbled upon this really cool new feature. I have to tell you about it, because this, imho, is one of the things that makes Oracle Clusterware a “Grid”.

So, what is Policy-based management? In order to answer that, you must first understand some basic changes in 11gR2. This is what the manual says:

With Oracle Clusterware 11g release 2 (11.2) and later, resources are contained in logical groups of servers called server pools. Resources are hosted on a shared infrastructure and are isolated with respect to their resource consumption by policies, behaving as if they were deployed in a single-system environment.

What does this mean? I have a bunch of resources and a bunch of servers (a grid) and subsets of these servers are combined in server pools. The resources are assigned to these server pools, but the server pool acts as if it were just one system. The isolation is done on a policy base. The whole idea is being able to limit the capacity that is assigned to a resource, and have a set of server pools to service those resources. BTW. It is possible to assign a server to multiple server pools.

This kind of management enables you to dynamically assign capacity to meet your policies and priorities, as well as to allocate (physical) resources by importance. Your critical applications will always have a minimum of resources at its disposal. It also gives you the opportunity to limit applications to use a set of servers.

How does it work? You just assign the minimum and maximum number of nodes to a resource, as well as the priority and Oracle Clusterware will do the rest.

The servers will get a number of attributes, like a name (basically the node name), the pools to which it belongs and the state it is in with some details.

How do Server Pools work? They are sets of servers, that service resources. What they will try to do is to spread the workload over the available servers in the pool. It is possible to limit a database to run in a specific server pool.

The top-level server pool is an exclusive server pool, which means that any given server can only be a part of one top-level server pool. It is impossible to be a member of multiple top-level server pools. This way, the grid or cluster is logically divided.

All servers that join the cluster/grid are automatically assigned to the Free Pool initially. This Free Pool is a server pool that is automatically created at installation. From here, servers can be assigned to other server pools. Another pool, the Generic Pool is also automatically generated, and all servers will be part of this pool too, in order to ensure compatibility with prior releases. Any pre-11gR2 database will need to run in this generic server pool.

The server pool is given an minimum and maximum number of nodes. As soon as the number of nodes decreases below the minimum value, Clusterware will automatically take necessary action and take servers from other pools to increase the number of servers to meet the required minimum number.

If this poses a problem for other server pools, decisions are made upon the importance of the pools. When importance is equal, it is only possible to take servers, if the number of servers is higher than the minimum value.

All of the above can be done completely dynamically, however, manual manipulation is still possible. You might think: "Where is my control?", but if you set it up carefully, it could save you a bundle. For example, when you configure Grid Plug and Play (another one of those new features), it is possible to have a database bring up a new instance in the server pool and automatically create new log threads, undo tablespaces, etc. Isn't that way cool?

No comments:

Post a Comment