Consolidation is very popular word today to save costs:

• buy one huge server with a lot of CPU, memory, huge storage, fast network

• put as many as possible databases on it

• utilize it as efficient as possible

But consolidation brings problem how to share resources between databases:

• CPU (this article is related only to this resource)

• storage

• memory

• network

As default an Oracle database can see/use all CPUs on a host. In 11.2G there is new option called “Instance Caging” which enables to limit number of CPUs which can be seen/used by a database.

Enabling Instance Caging

1. Manually set dynamic parameter cpu_count on instance level. As default this parameter is set to maximum number of cpu available on a hosts.

alter system set cpu_count=2;

2. Enable any resource manager plan that manages CPU. Most easy is to turn on default Oracle plan DEFAULT_PLAN.

alter system set resource_manager_plan=DEFAULT_PLAN;

Monitoring Instance Caging

Checking if Instance Caging is enabled

1.Ccheck if cpu_count is set

select value from v$parameter 
where name = 'cpu_count' 
  and (isdefault ='FALSE' or ismodified != 'FALSE');

2. Check if Resource Manager is on and managing CPU

select name from v$rsrc_plan 
where is_top_plan = 'TRUE' 
  and cpu_managed ='ON';

Provisioning CPU

There are generally 2 ways to distribute your CPU between databases

Partitioning – sum of all CPUs assigned to all your databases equals number of CPUs on a host. Databases don’t interfere with each other.

Over-provisioning – sum of all CPUs assigned to all your databases is higher that number of CPUs on a host. Databases can impact on each other performance.

Partitioning is recommended on production hosts

Over provisioning is recommended on less critical systems like development

Important note – Instance caging hasn’t got any impact on licensing model. You still need to pay for all CPUs on a host which is available for Oracle software.

Alternatives to Instance Caging

Alternative methods to control CPU usage are:

Hard partitioning – partition your hardware on many separate boxes. Install databases on separate boxes. Requires advanced hardware to be used – supported on Solaris, AIX, HP-unix. Disadvantages high costs, dedicated vendor, complex to managed.

O/S Workload Managers – many operating system delivers own software to manage CPU usage and other resources like AIX Workload Manager, the HP-UX Workload Manager, and the Solaris Resource. Disadvantages you need to learn of them, each can work different per O/S, not available on all platforms.

Virtualization – additional layer to partition hardware using software level. Disadvantage too complex versus Instance Caging and lose IO performance for databases.

Have a fun 🙂

Tomasz