Static Performance Prediction in PCASE: A Programming Environment for Parallel Supercomputers

Contention-sensitive static performance prediction for parallel distributed applications

Multi-agent networked linear dynamic systems have attracted the attention of researchers in power systems, intelligent transportation, and industrial automation. The agents might cooperatively optimize a global performance objective, resulting in social optimization, or try to satisfy their own selfish objectives using a noncooperative differential game. However, in these solutions, large volumes of data must be sent from system states to possibly distant control inputs, thus resulting in high cost of the underlying communication network. To enable economically viable communication, a game-theoretic framework is proposed under the communication cost, or sparsity, constraint, given by the number of communicating state/control input pairs. As this constraint tightens, the system transitions from dense to sparse communication, providing the tradeoff between dynamic system performance and information exchange. Moreover, using the proposed sparsity-constrained distributed social optimization and noncooperative game algorithms, we develop a method to allocate the costs of the communication infrastructure fairly and according to the agents' diverse needs for feedback and cooperation. Numerical results illustrate utilization of the proposed algorithms to enable and ensure economic fairness of wide-area control among power companies.

In this paper we study the problem of scheduling parallel loops at compile-time for a heterogeneous network of machines. We consider heterogeneity in three aspects of parallel programming: program, processor and network. A heterogeneous program has parallel loops with different amount of work in each iteration; heterogeneous processors have different speeds; and a heterogeneous network has different cost of communication between processors. We propose a simple yet comprehensive model for use in compiling for a network of processors, and develop compiler algorithms for generating optimal and sub-optimal schedules of loops for load balancing, communication optimizations and network contention. Experiments show that a significant improvement of performance is achieved using our techniques.

Many program optimizations require exact knowledge of the sets of array elements that are referenced in or that flow between statements or procedures. Some examples are array privatization, generation of communications in distributed memory machines, or compile-time optimization of cache behavior in hierarchical memory machines. Exact array region analysis is introduced in this article. These regions exactly represent the effects of statements and procedures upon array variables. To represent the flow of these data, we also introduce two new types of array region analyses: in and out regions. The intraprocedural propagation is presented, as well as a general linear framework for interprocedural analyses, which handles array reshapes. The intra- and inter-procedural propagation of array regions is implemented in pips, the interprocedural parallelizer of fortran programs developed at Ecole des mines de Paris.

Language constructs that support parallel computing are relatively well recognized at this point, with features such as parallel loops (optionally with reduction operators), divide-and-conquer parallelism, and general parallel blocks. But what language features would make distributed computing safer and more productive? Is it helpful to be able to specify on what node a computation should take place, and on what node data should reside, or is that overspecification? We don't normally expect a user of a parallel programming language to specify what core is used for a given iteration of a loop, nor which data should be moved into which core's cache. Generally the compiler and the run-time manage the allocation of cores, and the hardware worries about the cache. But in a distributed world, communication costs can easily outweigh computation costs in a poorly designed application. This panel will discuss various language features, some of which already exist to support parallel computing, and how they could be enhanced or generalized to support distributed computing safely and efficiently.

Emerging large-scale monitoring applications rely on continuous tracking of complex data-analysis queries over collections of massive, physically-distributed data streams. Thus, in addition to the space- and time-efficiency requirements of conventional stream processing (at each remote monitor site), effective solutions also need to guarantee communication efficiency (over the underlying communication network). The complexity of the monitored query adds to the difficulty of the problem -- this is especially true for nonlinear queries (e.g., joins), where no obvious solutions exist for distributing the monitor condition across sites. The recently proposed geometric method offers a generic methodology for splitting an arbitrary (non-linear) global threshold-monitoring task into a collection of local site constraints; still, the approach relies on maintaining the complete stream(s) at each site, thus raising serious efficiency concerns for massive data streams. In this paper, we propose novel algorithms for efficiently tracking a broad class of complex aggregate queries in such distributed-streams settings. Our tracking schemes rely on a novel combination of the geometric method with compact sketch summaries of local data streams, and maintain approximate answers with provable error guarantees, while optimizing space and processing costs at each remote site and communication cost across the network. One of our key technical insights for the effective use of the geometric method lies in exploiting a much lower-dimensional space for monitoring the sketch-based estimation query. Due to the complex, highly nonlinear nature of these estimates, efficiently monitoring the local geometric constraints poses challenging algorithmic issues for which we propose novel solutions. Experimental results on real-life data streams verify the effectiveness of our approach.

Genericity of parallel programming environments, enabling development of portable parallel programs, is expected to result in performance penalties. Furthermore, programmability and tool support of programming environments are important issues if a choice between programming environments has to be made. In this paper we propose a methodology to compare native and generic parallel programming environments, taking into account such competing issues as portability and performance. As a case study, this paper compares the Iserver-Occam, Parix, Express and PVM parallel programming environments on a 512-node Parasytec GCel. Furthermore, we apply our methodology to compare Parix and PVM on a new architecture, a 32-node Parsytec PowerXplorer, which is based on the PowerPC chip. In our approach we start with a representative application and isolate the basic (environment)-dependent building blocks. These basic building blocks, which depend on floating-point performance and communication capabilities of the environments, are analysed independently. We have measured point-to-point communication times, global communication times and floating-point performance. All information is combined into a time complexity analysis, allowing comparison of the environments on different degrees of functionality. Together with demands for portability of the code and development time (i.e. programmability), an overall judgement of the environments is given.

We consider the application of parallel programming environments to develop portable and efficient high performance data mining (DM) tools. We first assess the need of parallel and distributed DM applications, by pointing out the problems of scalability of some mining techniques and the need to mine large, eventually geographically distributed databases. We discuss the main issues of exploiting parallel and distributed computation for DM algorithms. A high-level programming language enhances the software engineering aspects of parallel DM, and it simplifies the problems of integration with existing sequential and parallel data management systems, thus leading to programming-efficient and high-performance implementations of applications. We describe a programming environment we have implemented that is based on the parallel skeleton model, and we examine the addition of object-like interfaces toward external libraries and system software layers. This kind of abstractions will be included in the forthcoming programming environment ASSIST. In the main part of the chapter, as a proof-of-concept we describe three well-known DM algorithms, Apriori, C4.5, and DBSCAN. For each problem, we explain the sequential algorithm and a structured parallel version, which is discussed and compared to parallel solutions found in the literature. We also discuss the potential gain in performance and expressiveness from the addition of external objects on the basis of the experiments we performed so far. We evaluate the approach with respect to performance results, design, and implementation considerations.

HPC the easy way: new technologies for high performance application development and deployment

A topology-aware load balancing algorithm for clustered hierarchical multi-core machines

S.213-220

Parkinson's Disease (PD) is a progressive neurodegenerative disorder characterized by motor dysfunction, cognitive decline, and various non-motor symptoms. Early detection and accurate monitoring of the disease remain critical for effective intervention and management. Recent advancements in machine learning (ML) have shown promise in identifying biomarkers that can assist in the diagnosis and progression tracking of Parkinson's Disease. This study explores the role of hierarchical machine learning models in decoding PD biomarkers, aiming to improve predictive accuracy and interpretability. Hierarchical models, which organize data processing in a multi-level structure, allow for better integration of complex, multi-dimensional biomedical data, such as neuroimaging, genetic, and clinical information. By employing techniques like deep learning, decision trees, and ensemble methods, hierarchical models can capture intricate relationships and patterns that traditional ML models often overlook. This research examines how hierarchical approaches enhance the identification of key PD biomarkers, including those associated with motor symptoms, neurodegeneration, and cognitive impairment. Furthermore, the study discusses the potential for these models to offer personalized predictions, paving the way for more tailored treatments. The results demonstrate that hierarchical machine learning models can significantly improve diagnostic accuracy, offering new insights into the molecular and physiological underpinnings of Parkinson's Disease.

Article Free Access Share on Static performance prediction of data-dependent programs Authors: Hasyim Gautama Faculty of Information Technology and Systems, Delft University of Technology, P.O. Box 5031, NL-2600 GA Delft, The Netherlands Faculty of Information Technology and Systems, Delft University of Technology, P.O. Box 5031, NL-2600 GA Delft, The NetherlandsView Profile , Arjan J. C. van Gemund Faculty of Information Technology and Systems, Delft University of Technology, P.O. Box 5031, NL-2600 GA Delft, The Netherlands Faculty of Information Technology and Systems, Delft University of Technology, P.O. Box 5031, NL-2600 GA Delft, The NetherlandsView Profile Authors Info & Claims WOSP '00: Proceedings of the 2nd international workshop on Software and performanceSeptember 2000 Pages 216–226https://doi.org/10.1145/350391.350435Online:01 September 2000Publication History 16citation249DownloadsMetricsTotal Citations16Total Downloads249Last 12 Months5Last 6 weeks1 Get Citation AlertsNew Citation Alert added!This alert has been successfully added and will be sent to:You will be notified whenever a record that you have chosen has been cited.To manage your alert preferences, click on the button below.Manage my Alerts New Citation Alert!Please log in to your account Save to BinderSave to BinderCreate a New BinderNameCancelCreateExport CitationPublisher SiteeReaderPDF

The authors discuss a concurrent programming environment and its application to a two-arm cooperative manipulation experiment on RIPS (robot instruction processing system). RIPS is a hierarchical multiprocessor architecture in which various custom and general-purpose processors are applied to a partitioning of the robot control problem. The system provides hardware support for synchronization and communication primitives, making it easier to write concurrent programs for RIPS' heterogeneous processors. The experiment demonstrated the viability of RIPS in supporting computationally intensive robot control methodologies, and as a byproduct has helped to develop a parallel programming environment for RIPS, called USE RIPS (user software environment for RIPS). By building USE RIPS on Unix and using a layered approach, it is possible to adapt or make use of various existing programs and utilities, such as the GNU and C++ compilers. >

Presents a framework of using resource metrics to characterize the various models of parallel computation. Our framework reflects the approach of recent models to abstract architectural details into several generic parameters, which we call resource metrics. We examine the different resource metrics chosen by different parallel models, categorizing the models into four classes: the basic synchronous models, and three extensions of the basic models which more accurately reflect practical machines by incorporating the notions of asynchrony, communication cost and memory hierarchy. We then present a new parallel computation model, the LogP-HMM model, as an illustration of design principles based on the framework of resource metrics. The LogP-HMM model extends an existing parameterized network model (LogP) with a sequential hierarchical memory model (HMM) characterizing each processor. The result accurately captures both network communication costs and the effects of multilevel memory, such as local cache and I/O. We examine the potential utility of our model in the design of near-optimal sorting and FFT algorithms. >