Context-Sensitive Load Balancing in Distributed Computing Systems

Jobs to be processed in distributed systems are often characterized by a high complexity. Several interdependent tasks (here called "steps") have to be performed on possibly different sites using different types of resources (services) [1,2,5-7]. As an example, think of a print job (job type J1) submitted by a user to be printed on some special printer. This print job consists of a database access step for the retrieval of data to be printed, a first transfer step via the communication network, a formatting step on a compute server, another transfer step transmitting the data to the print server, and the printing step itself. A look at this job in more detail will reveal even more steps involved. Different job types are processed concurrently in distributed computing systems. Steps of different job types may seek access to the same or to different servers. To extend the example above, a job of a different type (J2) to be processed concurrently could be a machine tool access job submitted via another terminal, retrieving data from a database and sending the data to a machine tool. After performing the requested manufacturing step, data collected during the execution for statistical evaluations may have to be written back to the database. With distributed systems growing in size, the number of servers providing the same service within the system is increasing. Therefore, a major issue is the choice of an adequate load sharing strategy, in order to reduce job response times and to improve resource utilization by distributing load among the servers. The problem of load sharing is well-known and several approaches to various aspects of this problem have been published in the past. The range of this problem is shown by the following extremes. On one hand, scheduling is an important issue in manufacturing and project management [5,6]. Here, the scheduling problem is treated as a large combinatorial problem. One definition of the scheduling problem in a flexible manufacturing environment is as follows: ’Given a set of production facilities and technological constraints, given a requirement, expressed in terms of quantity, product quality, and time constraints, find a feasible sequence of processing operations on the various facilities which satisfies the objectives’ [6]. However, scheduling in flexible, Abstract