Quality of Service Management for Enterprise-Application Computing Utilities

Suny At Stony Brook, Stony Brook NY

Investigators

Abstract

A growing concern among corporate information technology (IT) managers is the elevating gap between the increasing system maintenance and administration cost of the IT infrastructure and the decreasing hardware/software acquisition cost. This gap is escalating because system administration and management is still heavily labor-intensive and does not scale very well with system size and complexity, despite extensive research efforts to automate the process. One promising approach to drastically reduce this cost is to adopt a computing utility model, in which a centrally managed computing infrastructure is shared among multiple enterprise applications in a way that preserves their high-level business objectives. The primary technical challenge in the computing utility model is how to provision and schedule system resources to support the desired quality of service (QoS) of individual business applications that share the same physical infrastructure without human intervention. This project proposes to develop a multi-dimensional resource virtualization system called Iguacu, which multiplexes multiple system resources among a set of enterprise applications on a large-scale PC cluster such that the delay and/or throughput requirement of each such application is satisfied. Iguacu features three key innovations. First, unlike most previous QoS efforts that focus on a single system resource, Iguacu coordinates the allocation and scheduling of multiple system resources, including CPU, disk, and network link, to meet application-specific QoS requirements while significantly improving individual resource's utilization efficiency. Second, Iguacu supports a unified and reusable implementation framework that could quickly convert, with zero or minor modification, a given database-intensive enterprise application into a form that provides guarantees based on service-specific performance metrics. The development of such a framework greatly mitigates the efforts required to deploy QoS-aware enterprise applications in a complicated IT environment. Third, Iguacu features a novel procedure that automatically translates high-level QoS specifications into physical system resource requirements for distributed enterprise applications running on a cluster environment. Besides its intellectual contributions, this project is expected to make a major impact on the undergraduate and graduate education of the Computer Science department at Stony Brook University by introducing a series of system management and administration courses, which will use the Iguacu prototype as a case study on the functionality and implementation details of an enterprise-scale system management system for scalable PC clusters. Finally, the technologies developed in this project are expected to influence the future product directions of a major system management solution vendor, Computer Associates International, through an on-going research collaboration in the network/system management area.