Workload management dispatcher behavior in multimember environments

On multimember DB2 instances, the dispatcher behaves in the following ways:

• The relative quantity of CPU shares are evaluated and CPU resources are allocated across all members in the instance based on the work that is active in that instance on that host or LPAR as a whole. For example, let's consider a two member database with work being executed in two service classes, A and B. Service class A is assigned 3500 soft CPU shares and service class B is assigned 6500 soft CPU shares. The CPU utilization for the entire host or LPAR is 35% for service class A and 65% for service class B despite the possibility that relatively more work is being executed on one member than the other. The workload management dispatcher looks across both members when making its decisions regarding which service class should be getting more CPU resources at any given moment.
  Figure 1. Multimember database environments: Soft CPU shares allocations pie chart across 2 database members

  
• The value of the wlm_disp_concur database manager configuration parameter applies to each host or LPAR. In multimember databases, the dispatch concurrency level specified by the wlm_disp_concur configuration parameter is applied across all of the members.
• For multimember databases, CPU limits are enforced on a per host or LPAR basis - when multiple members are used, each partition is allowed to use all of a specified CPU limit as long as the sum across all members is less than or equal to the CPU limit. If the CPU limit on the single service class A is 35% and there are 2 members, then member 1 can use up to 35% if the sum of the CPU utilization across member 1 and 2 does not exceed 35%.
  Figure 2. Multimember database environments: CPU limit pie chart across two database members

  

The workload management dispatcher supports micropartition environments in AIX. In micropartition environments, the percentage CPU utilization for both controls and monitoring in the workload management dispatcher is computed relative to the CPU resources allocated to the LPAR by the operating system or hypervisor over the most recent dispatcher CPU resource scheduling cycle, using the entitled (guaranteed) CPU resources for the LPAR as a baseline level. In cases when the CPU resource allocated to the LPAR is greater than the entitled CPU resources, the dispatcher computes the CPU utilization relative to the greater amount, and when the CPU resource allocation is less than the entitled CPU resources, the dispatcher computes the CPU utilization relative to the entitled CPU resources.