Peter B. Yocom

Adaptive algorithms for managing a distributed data processing workload

Workload management, a function of the OS/390™ operating system base control program, allows installations to define business objectives for a clustered environment (Parallel Sysplex™ in OS/390). This business policy is expressed in terms that relate to business goals and importance, rather than the internal controls used by the operating system. OS/390 ensures that system resources are assigned to achieve the specified business objectives. This paper presents algorithms developed to simplify performance management, dynamically adjust computing resources, and balance work across parallel systems. We examine the types of data the algorithms require and the measurements that were devised to assess how well work is achieving customer-set goals. Two examples demonstrate how the algorithms adjust system resource allocations to enable a smooth adaptation to changing processing conditions. To the customer, these algorithms provide a single-system image to manage competing workloads running across multiple systems.

Intelligent resource director

Intelligent Resource Director (IRD), a feature of the IBM eServer zSeries™ processors and the z/OS™ operating system, manages multiple heterogeneous workloads with various business priorities toward achieving their goals. It establishes a more synergistic relationship with the Workload Manager (WLM) component of z/OS and the zSeries hardware, and augments the adjustments that WLM makes to local sysplex members by managing the set of logical partitions on a particular central processing complex (CPC) that are part of the same sysplex, known as a logically partitioned (LPAR) cluster. This paper describes the three primary areas that IRD manages: LPAR CPU management, channel subsystem priority queueing (CSSPQ), and dynamic channel path management (DCM).

IBM zEnterprise unified resource manager platform performance management

Large enterprise systems are composed of multiple applications deployed on sets of computing resources that compete for limited resources. Hybrid systems include a deployment of multiple hypervisors on different types of computing resources. Goal-based policies are deployed into these systems that indicate the relative business importance of the applications in the system, as well as the performance objectives of these applications. These policies are generally written without respect to the deployed architecture or hypervisor. The task of a goal-based performance management system is both to assign resources to applications in a manner that is consistent with the operating constraints that have been configured for the system and to assign platform resources to applications in a way that meets the policy objectives and makes tradeoffs in a manner consistent with the relative business importance of the applications. Such a management system has been developed for the IBM zEnterprise® hybrid architecture. This paper describes how goals and policies are defined for these systems and how the resources within zEnterprise are managed to meet the performance goals for each application in the context of what has been more specifically defined as a workload.