Intelligent fuzzy control to augment scheduling capabilities of network queuing systems

Abstract

Modern cost-effective computing in industry is heavily dependent on the ability of engineering management to replace older (more expensive) mainframes with the use of idle (less expensive) resources that are a part of the engineers normal work environment. With this shift in paradigm comes a need to duplicate many of the job management facilities that are normally associated with central mainframes. One example of this comes in the form of job scheduling. In particular, if engineers are to rely on clusters of workstations as a replacement to a mainframe for most of their routine work, then reliable job scheduling that ensures fault tolerance, interoperability of systems software, and load balancing is a necessary requirement. At present, most job scheduling software, that is available for the cluster environment, does not have the ability to schedule a job on a cluster based on network load. This work is an attempt at addressing some of the fundamental issues regarding network load balancing so that deployment requirements for network job scheduling of parallel and distributed computing applications can better be understood.

In this paper, a control system (via Fuzzy Logic) is described that prioritizes the allocation of parallel jobs to, and their suspension from, a cluster of networked workstations. The algorithm presented here augments the scheduling capabilities of existing network queuing systems by including, in addition to the available number of machines, network load as a cluster resource and an application's communication requirements as a resource specification in the ranking decisions. It will be demonstrated that the resulting fuzzy controller of this paper can be used as an easily extensible, inexpensive, modular interface to network queuing systems for the scheduling of parallel and sequential jobs.