A Performance Model for Disintegrated Manycores

A New Paradigm of Individual, Group, and Organizational Performance

The Unified Parallel C (UPC) programming language offers parallelism via logically partitioned shared memory, which typically spans physically disjoint memory subsystems. One convenient feature of UPC is its ability to automatically execute between-thread data movement, such that the entire content of a shared data array appears to be freely accessible by all the threads. The programmer friendliness, however, can come at the cost of substantial performance penalties. This is especially true when indirectly indexing the elements of a shared array, for which the induced between-thread data communication can be irregular and have a fine-grained pattern. In this paper, we study performance enhancement strategies specifically targeting such fine-grained irregular communication in UPC. Starting from explicit thread privatization, continuing with block-wise communication, and arriving at message condensing and consolidation, we obtained considerable performance improvement of UPC programs that originally require fine-grained irregular communication. Besides the performance enhancement strategies, the main contribution of the present paper is to propose performance models for the different scenarios, in the form of quantifiable formulas that hinge on the actual volumes of various data movements plus a small number of easily obtainable hardware characteristic parameters. These performance models help to verify the enhancements obtained, while also providing insightful predictions of similar parallel implementations, not limited to UPC, that also involve between-thread or between-process irregular communication. As a further validation, we also apply our performance modeling methodology and hardware characteristic parameters to an existing UPC code for solving a 2D heat equation on a uniform mesh.

It is our great pleasure to welcome you to the 22nd Workshop on Performance Engineering, Modelling, Analysis, and Visualization Strategy - PERMAVOST 2022. Modern software engineering is getting increasingly complicated. In the HPC field, we are dealing with cutting-edge infrastructure and a novel problem on an unprecedented scale. The scale here is not only in terms of the HPC infrastructure but also, in terms of the people involved. Modernday HPC projects involve a wide variety of experts. From application analysts, and tools developers to domain scientists, all with varying degrees of knowledge in HPC performance analysis. We created this workshop in the hope of bridging these stakeholders, creating a robust feedback loop to answer future challenges in performance modeling, analysis, engineering, and visualization.

It is our great pleasure to welcome you to the 1st Workshop on Performance Engineering, Modelling, Analysis, and Visualization Strategy - PERMAVOST 2021. Modern software engineering is getting increasingly complicated. In the HPC field, we are dealing with cutting-edge infrastructure and a novel problem with unprecedented scale. The scale here is not only in terms of the HPC infrastructure but also, in terms of the people involved. Modern-day HPC projects involve a wide variety of experts. From application analysts, tools developers to domain scientists, all with varying degrees of knowledge in HPC performance analysis. We created this workshop in the hope of bridging these stakeholders, creating a robust feedback loop to answer future challenges in performance modeling, analysis, engineering, and visualization.

The situation of telecommunications systems is rapidly evolving towards a state characterized by the presence of a multitude of services integrated in a single network structure. In order to ease their integration, it is important to provide an unambiguous description of the organization, of the functionality, and of the relationships existing among services. Since it is not possible to completely foresee the situation of services in the future, the description should be modular, so that the task of adding new services to an existing environment is simplified. Formal description techniques are languages designed to handle such descriptions. Their goal is to produce correct and unambiguous specifications, understanding and modeling the system and the domain within which it operates. Their input is a set of required statements and a conceptual overview of the system. Their output is a formal representation that captures the functional behavior. On the other hand, the throughput of physical networks is constantly growing, reaching the order of Gbit/s. In the meantime, services needing high throughput, like multimedia conferencing systems are being designed, creating a need for flexible, high performance communication systems. An important feature of new high speed services is their increased need for guarantees. Therefore, the use of powerful performance modeling techniques becomes of prime importance. Their goal is to obtain some notion of how the system will perform under a given set of conditions. Performance evaluation theorists have long been using system specifications to make a heuristic model of the system and then an analysis of the model. When system specifications exist, the question arises whether a more direct path to performance analysis can be found. Our scope is to provide a methodology that fully and consistently integrates formal specification and performance evaluation. A problem is that many of the bottlenecks and impediments to smooth data flow have to do with implementation issues not directly dealt within the system specification itself. Therefore, prior to generating a performance model, the formal specification should be enhanced with the relevant non-functional information. It is essential to assure a smooth specification of the non-functional information. Thus, it is necessary to consistently add this information to the formal specification. Besides, the extended formal method should focus on conserving the semantics of the original one, and on providing a coherent extension of the semantics and syntax for the new concepts. Since all relevant required information is integrated within the formal extended representation, we provide a mapping technique that translates the extended formal representation into a performance modeling conserving the properties of the formal model. In parallel to the development of a performance model, the formal model can be validated and verified, and their implementation can be tested for conformance towards the original formal model. Finally, the performance modelling is implemented in a programming language, and then simulated. The results from simulation can be reused as inputs for the optimization of the formal specification (functional optimization) or the extended formal model (design optimization). To sum up, the goal of this work is to: Define extensions to current formal methods that allow to enhance the formal specification with performance modeling related information. Develop a method to translate the extended formal specification into a performance modeling. Develop rules and guidelines to automatically implement the performance modeling for the purpose of a performance evaluation (we use the commercial tool Opnet). Develop a consistent and coherent methodology for the integration of performance modeling and formal specification.

The overriding aim of much product development research is to improve the performance of the Product Development Process (PDP). Yet the concept of performance as applied to this area is currently without underlying theory and no consensus exists on its definition or the most appropriate metrics to use in evaluation. This paper presents a review of research in PDP performance to illustrate the various issues that have received attention in this area. A definition and model of performance are proposed which are directed at developing a fundamental understanding of the phenomenon. The model illustrates the main elements of performance - efficiency and effectiveness - and provides a basis upon which to identify, formalise and analyse suitable performance metrics at any level or range in the product development process. Factors influencing performance have been assessed by the authors and sample results are presented. It is concluded that, although the model determines the main elements of performance, further research is necessary to truly formalise performance and specify a suitable approach for its measurement.

This article reports the development and validation of an integrated model of performance on a chemical concept ‐ volumetric analysis. From the chemical literature a path‐analytic model of performance on volumetric analysis calculation was postulated based on studies utilizing the proportional reasoning schema of Piaget and the Cumulative learning theory of Gagne. This integrated model hypothesized some relationships among the variables: direct proportional reasoning, inverse proportional reasoning, prerequisite concepts (content) and performance on volumetric analysis calculations. This model was postulated for the two groups of students involved in the study ‐ that is those who use algorithms with understanding and those who use algorithms without understanding. Two hundred and sixty‐five grade twelve chemistry students in eight schools (14 classes) in the lower mainland of British Columbia, Canada participated fully in the study. With the exception of the test on volumetric analysis calculations all the other tests were administered prior to the teaching of the unit on volumetric analysis. The results of the study indicate that for subjects using algorithms without understanding, their performance on VA problems is not influenced by proportional reasoning strategies while for those who use algorithms with understanding, their performance is influenced by proportional reasoning strategies.

Recently, the Lattice Boltzmann method is widely used in simulating fluid flows. In this paper, we present the performance analysis, modeling and prediction of a parallel multiblock Lattice Boltzmann application on up to 512 processors on three SMP clusters: two IBM SP systems at San Diego Supercomputing Center (DataStar ? p655 and p690) and one IBM SP system at the DOE National Energy Research Scientific Computing Center (Seaborg) using the Prophesy system. By characterizing the performance of the Lattice Boltzmann application as the problem size and the number of processors increase, we can identify and eliminate performance bottlenecks, and predict the application performance. The experimental results indicate that the application with large problem sizes scales well across these three clusters, and performance models using the coupling method are accurate with less than 4.8% average relative prediction error.

The adequacy of digital literacy (DL) among entrepreneurs can trigger the birth of readiness and ability to carry out digital transformation (DT) in relevant business processes and models in the digital economy era to improve entrepreneurial performance with better competitiveness. Also the existence of an entrepreneurial commitment to market orientation (MO) will facilitate business success. This research identifies and explores entrepreneurial performance (EP) models in the digital economy era in terms of DT, DL and MO mediation. The path analysis model was used to test the combined interaction of DT and DL with EP, and the MO relationship that mediates all three. The findings, DT and DL had a significant positive effect on MO both individually and jointly. DT directly has a significant positive effect on EP, also DL directly has a significant positive effect on EP. Sobel test results, it is known that DT has an insignificant positive effect on EP through MO, while DL has a significant positive effect on EP through MO. MO has the mediating effect of DT and DL relationships in influencing EP, consequently MO has a positive impact on EP. This research offers a framework for the performance model of entrepreneurship in a business perspective in the era of the digital economy. The adoption and modification of marketing mix factors into new marketing mix models is necessary to support the implementation of strategies and practices. An exploration of the influence of innovative new marketing mix models on DT practices, DL readiness and MO clarity in the context of entrepreneurship, as well as their implications for business performance, can be considered as future research directions.

Analytical (predictive) application performance models are critical for diagnosing performance-limiting resources, optimizing systems, and designing machines. Creating models, however, is difficult because they must be both accurate and concise. To ease the burden of performance modeling, we developed Palm (Performance and Architecture Lab Modeling tool), a modeling tool that combines top-down (human-provided) semantic insight with bottom-up static and dynamic analysis. First, Palm provides a source code modeling annotation language for abstracting or expressing complexity. Second, Palm generates hierarchical models according to well-defined rules. Since a model's hierarchy is defined by static and dynamic source code structure, there is a link between a program's organization and its model. By coordinating models and source code, Palm's models are 'first-class' and reproducible. Third, Palm incorporates measurements to focus attention, represent constant behavior, and validate models. We discuss generating models for three different applications.

Categorizing learning analytics models according to their goals and identifying their relevant components: A review of the learning analytics literature from 2011 to 2019

Distributed file systems are emerging as a key component of large scale cloud storage platform due to the continuous growth of the amount of application data. Performance modeling and analysis is an important concern in the distributed file system area. This paper focuses on the performance prediction and modeling issues. An adaptive prediction model (APModel) is proposed to predict the performance of distributed file systems by capturing the performance correlation of different performance factors. We perform a series of experiments to validate the proposed prediction model. The experiment results indicate our proposed approach can get better prediction accuracy. It is practical and can achieve sufficient performance analysis for distributed file systems.

A new performance and performability modeling tool is introduced in this paper. PENELOPE is the first tool which incorporates evaluation and optimization algorithms. It is the result of a combination between the performability modeling concept and Markov decision theory. Different algorithms are adopted and included in the tool under the unifying paradigm of reconfigurability as the basis for adaptation and optimization. In addition to transient and steady-state performability measures, also transient and stationary control functions can be computed and graphically presented. Model specification and specification of transient or stationary control functions can be separately performed and deliberately combined with each other. Besides providing a new modeling paradigm, the tool supports model creation, experimentation, storage and presentation of results by means of an easily usable interface and an integrated model data base system.KeywordsArrival RateService RateCongestion ControlReward RateSwitching CurveThese keywords were added by machine and not by the authors. This process is experimental and the keywords may be updated as the learning algorithm improves.

The detailed design and development phase of the X-SAR 9.6-GHz vertical polarization segment of the Shuttle Radar Lab (SRL) mission, started in Spring 1987, is discussed. The design incorporates a planar array antenna based on metallized carbon-fiber-reinforced plastic technology, a traveling-wave-tube amplifier, and a dual-channel receiver using surface acoustic wave devices. Several radar parameters are controllable from ground to achieve high performance over a wide range of measurement geometries. Definitions of image performance parameters and modeling algorithms developed in support of instrument specification and performance monitoring during hardware development are presented. The algorithms have been implemented as a software tool to perform sensitivity analysis and generate interesting graphical results to support top-level instrument verification.< <ETX xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">&gt;</ETX>

Design-time performance modeling of compositional parallel programs