Towards Autonomic Workload Management in DBMSs

Introduction

Workload management involves the monitoring and control of work entering a system. Its goal is to ensure that sufficient resources are allocated to a workload so that a business can meet its management objectives (IBM Corporation, 2003a). Workload management is becoming increasingly important to businesses for several reasons. First, their workloads are becoming more diverse and complex. Second, the emerging trend of server consolidation has led to an environment with increased competition for shared resources between applications from potentially disjointed organizations, which results in a workload with diverse and dynamic resource demands and often competing performance objectives. Third, Web-based applications, especially Web services (Erickson & Siau, 2008), introduce a need for flexible and guaranteed application service levels because they tend to involve unpredictable workloads, with a high rate of overall growth in workload size (D. H. Brown Associate, Inc, 2004). Allocating DBMS resources to competing workloads to meet performance objectives is a challenge.

We believe that workload management stands to benefit greatly from the new paradigm of autonomic computing. The goal of autonomic computing is to simplify system complexity by governing all computations within a given system (Mainsah, 2002). Autonomic computing systems have four fundamental characteristics (Ganek & Corbi, 2003): self-configuring, self-healing, self-optimizing and self-protecting. Self-configuring means systems can automatically adapt to dynamically changing environments. New features, applications and servers can be dynamically added to the enterprise infrastructure with no disruption of services. Self-healing is the ability of systems to discover, diagnose and react to disruptions. Such a system needs to be able to predict problems and take necessary actions to prevent the failures from impacting the services. For a system to be self-optimizing, it should monitor itself and tune resources automatically to maximize resource utilization to meet users’ performance requirements. Self-protecting means systems are able to anticipate, detect, identify, and protect themselves from attacks from anywhere.

The goal of this paper is to specifically examine the progress made towards providing autonomic workload management in DBMSs and, based on this examination, to identify directions for future research. Previous surveys of the autonomic computing area, such as those by Salehie and Tahvildari (2005) and Elnaffar, Powley, Benoit, and Martin (2003), tend to have a more general focus and use criteria focusing on the existence of facilities to support the general features of self-configuration, self-healing, self-optimization and self-protection. We propose a framework, Autonomic Workload Management Framework (AWMF), for the examination of autonomic workload management and then analyze current products and research efforts with respect to it. The framework specifies a model of the key processes and functions in autonomic workload management as well as identifying criteria for evaluation.

The remainder of the paper is structured as follows. The following section discusses the objectives of workload management. We then outline our framework, AWMF, for autonomic workload management and examine workload management mechanisms taken from current DBMS products and research efforts and compare them using AWMF. A number of observations are derived from the comparison of the workload management mechanisms. We then validate AWMF using experiments with Query Scheduler, a prototype implementation of AWMF and finally summarize our survey and point out possible directions for future work.