Abstract-One measure of usefulness of a general-purpose distrib-uted computing system is the system’s ability to provide a level of per-formance commensurate to the degree of multiplicity of resources pres-ent in the system. Many different approaches and metrics of performance have been proposed in an attempt to achieve this goal in existing systems. In addition, analogous problem formulations exist in other fields such as control theory, operations research, and produc-tion management. However, due to the wide variety of approaches to this problem, it is difficult to meaningfully compare different systems since there is no uniform means for qualitatively or quantitatively eval-uating them. It is difficult to successfully build upon existing work or identify areas worthy of additional effort without some understanding of the relationships between past efforts. In this paper, a taxonomy of approaches to the resource management problem is presented in an attempt to provide a common terminology and classification mecha-nism necessary in addressing this problem. The taxonomy, while pre-sented and discussed in terms of distributed scheduling, is also appli-cable to most types of resource management. As an illustration of the usefulness of the taxonomy an annotated bibliography is given which classifies a large number of distributed scheduling approaches accord-ing to the taxonomy. Index Terms-Distributed operating systems, distributed resource management, general-purpose distributed computing systems, sched-uling, task allocation, taxonomy. T I.

It is now more than a quarter of a century since researchers started publishing papers on mapping strategies for distributing computation across the computation resource of multiprocessor systems. There exists a large body of literature on the subject, but there is no commonly-accepted framework whereby results in the field can be compared. Nor is it always easy to assess the relevance of a new result to a particular problem. Furthermore, changes in parallel computing technology have made some of the earlier work of less relevance to current multiprocessor systems. Versions of the mapping problem are classified, and research in the field is considered in terms of its relevance to the problem of programming currently available hardware in the form of a distributed memory multiple instruction stream multiple data stream computer: a multicomputer.

he development of operating systems for parallel computers has closely followed that for serial computers. At first, the most advanced parallel computers ran in batch mode or single-user mode. At best, they allowed a static partitioning among a number of users. They were typically designed with a specific computational task in mind or for a certain class of computations. Like serial computers, they are currently evolving towards general-purpose, interactive, multiuser parallel systems. To explain the underlying motivation for our work, we note that a general-purpose, interactive, multiuser, multiprogramming parallel environment has the following advantages (in addition to the traditional advantages in uniprocessor environments, such as cost effectiveness): l This environment provides users with a spectrum of computational powers, cov-ering the range from personal computers to supercomputers. A user requiring more computational power can simply use more processors. Thus, a short response time for both simple and computationally intensive tasks is possible. l The spectrum of powers also aids pro-gram development and evaluation. Ini-tially, only one processor is needed. Addi-tional processors can be added later with-

Interprocessor communication (PC) overheads have emerged as the major performance limitation in parallel processing systems, due to the transmission delays, synchronization overheads, and conflicts for shared communication resources created by data exchange. Accounting for these overheads is essential for attaining efficient hardware utilization. This thesis introduces two new compile-time heuristics for scheduling precedence graphs onto multiprocessor architectures, which account for interprocessor communication overheads and interconnection constraints in the architecture. These algorithms perform scheduling and routing simultaneously to account for irregular interprocessor interconnections, and schedule all communications as well as all computations to eliminate shared resource contention. The first technique, called dynamic-level scheduling, modifies the classical HLFET list scheduling strategy to account for IPC and synchronization overheads. By using dynamically changing priorities to match nodes and processors at each step, this technique attains an equitable tradeoff between load balancing and interprocessor communication cost. This method is fast, flexible, widely targetable, and displays promising perforrnance. The second technique, called declustering, establishes a parallelism hierarchy upon the precedence graph using graph-analysis techniques which explicitly address the tradeoff between exploiting parallelism and incurring communication cost. By systematically decomposing this hierarchy, the declustering process exposes parallelism instances in order of importance, assuring efficient use of the available processing resources. In contrast with traditional clustering schemes, this technique can adjust the level of cluster granularity to suit the characteristics of the specified architecture, leading to a more effective solution.

A process is an operating system abstraction representing an instance of a running computer program. Process migration is the act of transferring a process between two machines during its execution. Several implementations

Abstract not found

ABSTRACT We present an optimal solution to the problem of allocating communicating periodic tasks to heterogeneous processing nodes (PNs) in a distributed real-time system. The solution is optimal in the sense of minimizing the maximum normalized task response time, called the system hazard, subject to the precedence constraints resulting from intercommunication among the tasks to be allocated. Minimization of the system hazard ensures that the solution algorithm will allocate tasks so as to meet all task deadlines under an optimal schedule, whenever such an allocation exists. The task system is modeled with a task graph (TG), in which computation and communication modules, communication delays, and intertask precedence constraints are clearly described. Tasks described by this TG are assigned to PNs by using a branch-and-bound (B&B) search algorithm. The algorithm traverses a search tree whose leaves correspond to potential solutions to the task allocation problem. We use a bounding method that prunes, in polynomial time, non-leaf vertices that cannot lead to an optimal solution, while ensuring that the search path leading to an optimal solution will never be pruned. For each generated leaf vertex we compute the exact cost using the algorithm developed in [1]. The lowest-cost leaf vertex (one with the least system hazard) represents an optimal task allocation. Computational experiences and examples are provided to demonstrate the concept, utility, and power of the proposed approach.

An efficient recursive task allocation scheme, based on the Kernighan-Lin mincut bisection heuristic, is proposed for the effective mapping of tasks of a parallel program onto a hypercube parallel computer. It is evaluated by comparison with an adaptive, scaled simulated annealing method. The recursive allocation scheme is shown to be effective on a number of large test task graphs -- its solution quality is nearly as good as that produced by simulated annealing, and its computation time is several orders of magnitude less.

Abstract-Optimal load allocation for load sharing a divisible job over processors interconnected in either a bus or a tree network is considered. The processors are either equipped with front-end processors or not so equipped. Closed form solutions for the minimum finish time and the optimal data allocation for each processor are obtained. The performance of large symmetric tree networks is examined by aggregating the component links and processors into a single equivalent processor. This allows an easy examination of large tree networks. In addition, it becomes possible to find a closed form solution for the optimal amount of data that is to be assigned to each processor in the tree network in order to achieve the minimum finish time. Index Terms- Load sharing, load balancing, divisible job, multiprocessors. I.

This paper presents MULTIFIT-COM, a static task allocator which could be incorporated into an automated compiler/linker/loader for distributed processing systems. The allocator uses performance information for the processes making up the system in order to determine an appropriate mapping of tasks onto processors. It uses several heuristic extensions of the MULTIFIT bin-packing algorithm to find an allocation that will offer a high system throughput, taking into account the expected execution and interprocessor communication requirements of the software on the given hardware architecture. Throughput is evaluated by an asymptotic bound for saturated conditions and under an assumption that only processing resources are required. A set of options are proposed for each of the allocator's major steps. An evaluation was made on 680 small randomly generated examples. Using all the search options, an average performance difference of just over 1% was obtained. Using a carefully chosen small su...