The invention provides a system and method for managing clusters of parallel processors for use by groups and individuals requiring supercomputer level computational power. A Beowulf cluster provides supercomputer level processing power. Unlike a traditional Beowulf cluster; however, cluster size in not singular or static. As jobs are received from users/customers, a Resource Management System (RMS) dynamically configures and reconfigures the available nodes in the system into clusters of the appropriate sizes to process the jobs. Depending on the overall size of the system, many users may have simultaneous access to supercomputer level computational processing. Users are preferably billed based on the time for completion with faster times demanding higher fees.

Dynamically allocated cluster system

Memory latency consideration for load sharing on heterogeneous network of workstations

Dynamic job processing based on estimated completion time and specified tolerance time

Loop partitioning on parallel and distributed systems has been a critical problem. Furthermore, it becomes more difficult to deal with on the emerging heterogeneous PC cluster environments. In the past, some loop self-scheduling schemes have been proposed to be applicable to heterogeneous cluster environments. In this paper, we propose a performance-based approach, which partitions loop iterations according to the performance ratio of cluster nodes. To verify the proposed approach, a heterogeneous cluster is built, and three types of application programs are implemented to be executed in this testbed. Experimental results show that the proposed approach performs better than traditional schemes.

/pdf/dynamic-partitioning-of-loop-iterations-on-heterogeneous-pc-kzvp4n7zoq.pdf

Dynamic partitioning of loop iterations on heterogeneous PC clusters

Recently, a series of parallel loop self-scheduling schemes have been proposed, especially for heterogeneous cluster systems. However, they employed the MPI programming model to construct the applications without considering whether the computing node is multicore architecture or not. As a result, every processor core has to communicate directly with the master node for requesting new tasks no matter the fact that the processor cores on the same node can communicate with each other through the underlying shared memory. To address the problem of higher communication overhead, in this paper we propose to adopt hybrid MPI and OpenMP programming model to design two-level parallel loop self-scheduling schemes. In the first level, each computing node runs an MPI process for inter-node communications. In the second level, each processor core runs an OpenMP thread to execute the iterations assigned for its resident node. Experimental results show that our method outperforms the previous works.

Using hybrid MPI and OpenMP programming to optimize communications in parallel loop self-scheduling schemes for multicore PC clusters

The Beowulf parallel-processing architecture is a public-domain software technology developed by NASA to link COTS processors yielding supercomputer performance (10 Gflops) very economically (under $150 K). This paper describes our implementation of a Beowulf cluster comprising 18 Sun workstations and the Linux operating system. We describe and demonstrate ParaSort, a distributed, parallel, data-allocation-sorting algorithm with automatic distributed load-balancing, and fault-tolerant performance. The load-balancing feature can provide dynamic, on-board, adaptive optimal distribution of processing tasks across a heterogeneous network of computing devices. Furthermore, it can intelligently allocate processing to off-board resources as appropriate and as they become available. We analyze the strengths and weaknesses of Beowulf and describe future extensions of our work.

W. Wren

Papers

Beowulf parallel processing for dynamic load-balancing