Analysis of Load Average and its Relationship to Program Run Time on Networks of Workstations

Recent work in high-performance computing has shifted attention to PC cluster. For PC-clusters, member nodes are independent computers connected by general-purpose networks. The latency of data communications is long and load balancing among the nodes becomes a critical issue. We introduce a new model for program development on PC clusters, namely the super-programming model (SPM) to address this issue. In SPM, PC clusters are modeled as a single virtual machine with PC as their processing units. The workload is modeled as a collection of super-instructions (SI). Each SI can achieve a limited workload. Application programs are coded using SI. SI is dynamically assigned to available PC at run time. For limited workload, no SI overloads any PC. Therefore, dynamic load balancing becomes an easier task. We apply SPM to mining association rules. Our experiments show that under normal conditions the workload is balanced very well. A performance model is also developed to describe the scalable behavior of SPM.

Dominant trends in parallel and distributed computing (parallel computers, parallel algorithms) aim to connect a number of powerful workstations (WS) based on single personal computers (PC) or symmetrical multiprocessor system (SMP) to solve complex problems in a parallel way. Typical representatives of these trends are network of workstation (NOW) and Grid as a high integrated network of NOW modules. In relation to it there is a bad need of unified modelling in traditionally evolved parallel computing and distributed computing as two separate research disciplines. Likewise current trends in high performance computing (HPC) aim to use NOW modules based on SMP parallel computers as cheaper alternatives to the traditionally used massive parallel multiprocessors or supercomputers and to profit from unifying modelling in parallel and distributed computing too. To exploit the parallel processing capability of such cluster, any application program must be paralleled. A way how to do it (decomposition model) belongs to a most important step in developing process of effective parallel algorithm. This article at first discusses the development of effective (optimised) iterative parallel algorithms (IPA) to solve partial differential equations (PDE). By carefully choosing an illustrative typical problem of PDE (Laplace equation) the proposed research has demonstrated the most important complexities in the process of complete modelling of substantial complexities and their optimisation in order to develop an effective PA. Such approach could be used as illustrative example for a complete modelling and effective solution of even other more complex PDE´s in a parallel way.

In this paper, the role of optical interconnects in high performance computing systems is explored. A Network-of- Workstations (NOW) consists of a collection of commodity workstations or high-end Personal Computers (PCs) interconnected with a high bandwidth Local Area Network (LAN). Due to technological advances, it has been argued that NOWs can offer cost-effective high performance computing systems, ranging from low-end interactive computing to high-end parallel computing. Over the last decade the industry has been moving in this direction: Commercial high end computing systems such as the Silicon Graphics Origin 2000 multiprocessor support up to 512 dual-processor nodes interconnected with a scalable network. In this paper, we argue that when NOWs are enhanced with a new generation of very high bandwidth optical networks, they can provide supercomputer-class performance at a fraction of the cost. We describe the architectures of ring-based and star-based multi- terabit optical networks for multiprocessor systems, and discuss their impact on system performance. These networks exploit the CMOS/VCSEL optoelectronic integrated circuit technology to yield exceptionally large bandwidths with small form factors and potentially low cost. We view these new optical networks as the key to the future evolution of multiprocessor systems, and we believe that such systems can open a new era of exceptionally high performance computing.

Task scheduling in real-time systems is a concept by which we can schedule the tasks according to their priorities. In hard real-time systems, the task scheduling is more important due to the chance of system failure. We discuss EDF task scheduling which is an optimal technique for the uniprocessor systems. The existing result of the QPA algorithm was applied on the uniprocessor system to check the schedulability of the tasks in uniprocessor system. We have extended the QPA algorithm to check the schedulability of the task sets in multiprocessor systems. Our algorithm is based on QPA algorithm and the utilization balanced algorithm. The proposed algorithm checks the absolute deadlines much faster than the existing QPA algorithm for uniprocessor real-time systems.

he analysis of any hardware or software from the point of view of the actual and possible performance is crucial in the computer world. Huge efforts are made to create newer, better and faster computers and programs while minor design flaws can result in very low performance not even worth the invested money. Many scientists focus their research on the performance analysis of different areas like parallel programs, PC clusters, shared memory architectures, supercomputers, cache performance and so on. Basically, two approaches can be found in this effort. One group of researchers try to create analitical models to calculate the performance and limits “in theory” while the other group implements monitoring tools to collect as much data about real executions as they can and then process this data and find bottlenecks in the performance. The first session of the workshop deals with performance analysis issues of different areas in computer science. The performance of PC clusters is analysed here as well as parallel applications with the need of the power of a teraflop supercomputer. Simulator for low-level interconnection networks is presented as well as mathematical models for parallel programs. Every researcher aims at making possible to the users to improve the performance of their system and their applications. The first paper is “Beowulf Performance in CFD Multigrid Applications” authored by F. Tirado, C. Garcia, R.S. Montereo, M. Prieto and I.M. Llorente. Authors evaluated the performance of a Beowulf-class system from the viewpoint of Computational Fluid Dynamics applications. They used two multigrid solvers to test their PC cluster with different configurations. The results show that, not surprisingly, the best configuration depends on the target application. The second paper presents “SICOSYS: An Integrated Framework for studying Interconnection Network Performance in Multiprocessor Systems” written by V. Puente, J.A. Gregorio and R. Beivide. SICOSYS is a generalpurpose simulator that makes possible the analysis of the interconnection networks of multiprocessor systems. This simulator has been integrated into the RSIM shared-memory-architecture and processor simulator as well as into the SimOS simulator that also includes the effects of an operating system into the simulation. The third paper presents new ideas about “Performance Analysis for Teraflop Computers: A Distributed Automatic Approach” authored by M. Gerndt, A. Schmidt, M. Schulz and R. Wismuller. A new performance analysis system has been designed for the monitoring and analysis of extremely computationally intensive applications on a teraflop computer. In such environment, classical analysis techniques fail because they are not scalable for thousands of processors. This paper presents the design of a new, automatic performance tool that is scalable on teraflop computers using a distributed approach for the analysis process. The fourth paper is the “Execution Time Prediction for Parallel Data Processing Tasks” written by S. Juhasz. This paper presents a mathematical model for execution time prediction in the domain of parallel programs with moderate computational power needs but working on large amount of data. A parallel sorting application is then analysed as an example.

The architecture of an automatically tuned linear algebra library proposed in previous works is extended in order to adapt itself to platforms where both the CPU load and the network traffic vary. During the installation process in a system, the linear algebra routines will be tuned automatically to the system conditions: hardware characteristics and basic libraries used in the linear algebra routines. At run-time the parameters that define the system characteristics are adjusted to the actual load of the platform. The design methodology is analysed with a block LU factorisation. Variants for sequential and parallel versions of this routine on a logical rectangular mesh of processors are considered. The behavior of the algorithm is studied with message-passing, using MPI on a cluster of PCs. The experiments show that the configurable parameters of the linear algebra routines can be adjusted during the run-time process despite the variability of the environment.

Parallel computing is an important area of research. It is used from scientific to commercial applications. Task scheduling is one of the important research areas of parallel computing. The major objective of any task scheduling algorithm is to minimize the execution time i.e. to reduce the scheduling length. Here, an application program is represented by a directed acyclic graph (DAG). In this paper, we have proposed an efficient CPTL scheduling algorithm which is based on two well-known attributes: Critical Path (CP), and Top-level (T-level). This algorithm is calculated the difference between CP and t-level, which is called CPT attribute. This attribute used as a priority of the tasks, and allocated these tasks on multiprocessor system. The proposed algorithm outperforms over the existing heuristic algorithms such as HLEFT, MCP, ETF, and DLS. CPTL algorithm gives minimum scheduling length. The comparison has been done based on following metrics: scheduling length, speedup, efficiency, load balancing, and normalized scheduling length.

In this paper, we will present the concept of the software system OpTiX for nonlinear, single and multiple criteria optimization, and our experiences with it. The concept is to use the “WYSIWYG” (What You See Is What You Get) approach from desktop publishing in combination with features from the object-oriented computer language paradigm. The system combines symbolic processing with numeric methods. The combination of several optimization methods (scanning, random search, direct search, sequential quadratic programming, generalized reduced gradients) assures a robust solution of a broad class of nonlinear optimization problems. The system as presented here was understood as a testbed for applying die above mentioned algorithms sequential and quasi-parallel to optimization tasks. These studies served as an early stage of research into using parallel and distributed processing (on multiprocessor systems with distributed memory or on networks of workstations) for the solution of optimization problems. For the analysis of multiple criteria problems the idea of the reference point method was used.

Parallel systems provide a robust approach for high performance computing. Lately the use of parallel computing has become more available as new parallel environments have evolved. Low cost and high performance of off-the-shelf PC processors have made PC-based multiprocessor systems popular. These systems typically contain two or four processors. Standardized POSIX-threads have formed an environment for the effective utilization of several processors. Moreover, distributed computing using networks of workstations has increased. The motivation for this work is to apply these techniques in computer vision. The Hough Transform (HT) is a well-known method for detecting global features in digital images. However, in practice, the sequential HT is a slow method with large images. We study the behavior of line detecting HT with both message passing workstation networks and shared-memory, multiprocessor systems. Parallel approaches suggested in this paper seem to decrease the computation time of HT significantly. Thus, the methods are useful for real-world applications.

Current compilers lack precise timing models guiding their built-in optimizations. Hence, compilers apply ad-hoc heuristics during optimization to improve code quality. One of the most important optimizations is register allocation. Many compilers heuristically decide when and where to spill a register to memory, without having a clear understanding of the impact of such spill code on a program's run time.

Loop fusion combines corresponding iterations of different loops, ft is traditionally used to decrease program run time, by reducing loop overhead and increasing data locality. In this paper, however, we consider its effect on energy. By merging program phases, fusion tends to increase the uniformity, or balance of demand for system resources. On a conventional superscalar processor, increased balance tends to increase IPC, and thus dynamic power, so that fusion-induced improvements in program energy are slightly smaller than improvements in program run time. If IPC is held constant, however, by reducing frequency and voltage - particularly on a processor with multiple clock domains - then energy improvements may significantly exceed run time improvements. We demonstrate the benefits of increased program balance under a theoretical model of processor energy consumption. We then evaluate the benefits of fusion empirically on synthetic and real-world benchmarks, using our existing loop-fusing compiler and a heavily modified version of the SimpleScalar/Wattch simulator. For the real-world benchmarks, we demonstrate energy savings ranging from 7-40%, with run times ranging from 1% slowdown to 17% speedup. In addition to validating our theoretical model, the simulation results allow us to tease apart the factors that contribute to fusion-induced time and energy savings.

Network of workstations (NOW) is a cost-effective alternative to a multiprocessor system. Here we propose a centralized architecture for parallel query processing on network of workstations. We describe a three-level processing strategy and evaluate its performance. The top two levels use a space-sharing technique to assign a partition to a query. The third-level uses a chunk-based load sharing policy to accommodate dynamic local load present on the workstations. To evaluate the performance of this three-level strategy, we have implemented it using PVM on a Pentium-based NOW.

In a Linux cluster, as in any multiprocessor system, the inter-processor communication rate is the major limiting factor to its general usefulness. This research is geared toward improving the communication performance by identifying where the inefficiencies lie and trying to understand their cause. The NetPIPE utility is being used to compare the latency and throughput of all current message-passing libraries and the native software layers they run upon for a variety of hardware configurations.

Communication-aware task scheduling and voltage selection for total energy minimization in a multiprocessor system using Ant Colony Optimization

Static task scheduling in multiprocessor frameworks is one of the well-defined NP hard problem. Due to optimal utilization of processors and in addition investing less time, the scheduling of tasks in multiprocessor frameworks is of extraordinary significance. To solve NP hard problem using traditional strategies takes reasonable measures of time. Over the time, various heuristic procedures were presented for comprehending it. Therefore, heuristic methods such as genetic algorithms are appropriate methods for task scheduling in multiprocessor system. In this paper, a new GA for static task scheduling in multiprocessor systems has been presented whose priority of tasks’ execution is based on the height of task in graph and other mentioned parameters and then scheduling is performed. This proposed method is simulated and then compared with basic genetic algorithm.