Alfredo Gimenez

State of the Art of Performance Visualization

Performance visualization comprises techniques that aid developers and analysts in improving the time and energy efficiency of their software. In this work, we discuss performance as it relates to visualization and survey existing approaches in performance visualization. We present an overview of what types of performance data can be collected and a categorization of the types of goals that performance visualization techniques can address. We develop a taxonomy for the contexts in which different performance visualizations reside and describe the state of the art research pertaining to each. Finally, we discuss unaddressed and future challenges in performance visualization.

Dissecting on-node memory access performance: a semantic approach

Optimizing memory access is critical for performance and power efficiency. CPU manufacturers have developed sampling-based performance measurement units (PMUs) that report precise costs of memory accesses at specific addresses. However, this data is too low-level to be meaningfully interpreted and contains an excessive amount of irrelevant or uninteresting information. We have developed a method to gather fine-grained memory access performance data for specific data objects and regions of code with low overhead and attribute semantic information to the sampled memory accesses. This information provides the context necessary to more effectively interpret the data. We have developed a tool that performs this sampling and attribution and used the tool to discover and diagnose performance problems in real-world applications. Our techniques provide useful insight into the memory behaviour of applications and allow programmers to understand the performance ramifications of key design decisions: domain decomposition, multi-threading, and data motion within distributed memory systems.

A flexible low-complexity device adaptation approach for data presentation

Visual data presentations require adaptation for appropriate display on a viewing device that is limited in re- sources such as computing power, screen estate, and/or bandwidth. Due to the complexity of suitable adaptation, the few proposed solutions available are either too resource-intensive or in exible to be applied broadly. Eective use and acceptance of data visualization on constrained viewing devices require adaptation approaches that are tailored to the requirements of the user and the capabilities of the viewing device. We propose a predictive device adaptation approach that takes advantage of progressive data renement. The approach relies on hierarchical data structures that are created once and used multiple times. By incrementally reconstructing the visual presentation on the client with increasing levels of detail and resource utilization, we can determine when to truncate the renement of detail so as to use the resources of the device to their full capacities. To determine when to nish the renement for a particular device, we introduce a prole-based strategy which also considers user preferences. We discuss the whole adaptation process from the storage of the data into a scalable structure to the presentation on the respective viewing device. This particular implementation is shown for two common data visualization methods, and empirical results we obtained from our experiments are presented and discussed.

Using R-trees for interactive visualization of large multidimensional datasets

Large, multidimensional datasets are difficult to visualize and analyze. Visualization interfaces are constrained in resolution and dimension, so cluttering and problems of projecting many dimensions into the available low dimensions are inherent. Methods of real-time interaction facilitate analysis, but often these are not available due to the computational complexity required to use them. By organizing the dataset into a level-of-detail (LOD) hierarchy, our proposed method solves problems of both inefficient interaction and visual cluttering. We do this by introducing an implementation of R-trees for large multidimensional datasets. We introduce several useful methods for interaction, by queries and refinement, to explain the relevance of interaction and show that it can be done efficiently with R-trees.We examine the applicability of hierarchical parallel coordinates to datasets organized within an R-tree, and build upon previous work in hierarchical star coordinates to introduce a novel method for visualizing bounding hyperboxes of internal R-tree nodes. Finally, we examine two datasets using our proposed method and present and discuss results.

A scalable observation system for introspection and in situ analytics

SOS is a new model for the online in situ characterization and analysis of complex high-performance computing applications. SOS employs a data framework with distributed information management and structured query and access capabilities. The primary design objectives of SOS are flexibility, scalability, and programmability. SOS provides a complete framework that can be configured with and used directly by an application, allowing for a detailed workflow analysis of scientific applications. This paper describes the model of SOS and the experiments used to validate and explore the performance characteristics of its implementation in SOSflow. Experimental results demonstrate that SOS is capable of observation, introspection, feedback and control of complex high-performance applications, and that it has desirable scaling properties.

Bootstrapping Parameter Space Exploration for Fast Tuning

The task of tuning parameters for optimizing performance or other metrics of interest such as energy, variability, etc. can be resource and time consuming. Presence of a large parameter space makes a comprehensive exploration infeasible. In this paper, we propose a novel bootstrap scheme, called GEIST, for parameter space exploration to find performance-optimizing configurations quickly. Our scheme represents the parameter space as a graph whose connectivity guides information propagation from known configurations. Guided by the predictions of a semi-supervised learning method over the parameter graph, GEIST is able to adaptively sample and find desirable configurations using limited results from experiments. We show the effectiveness of GEIST for selecting application input options, compiler flags, and runtime/system settings for several parallel codes including LULESH, Kripke, Hypre, and OpenAtom.

DR-BW: Identifying Bandwidth Contention in NUMA Architectures with Supervised Learning

Non-Uniform Memory Access (NUMA) architectures are widely used in mainstream multi-socket computer systems to scale memory bandwidth. Without a NUMA-aware design, programs can suffer from significant performance degradation due to inter-socket bandwidth contention. However, identifying bandwidth contention is challenging. Existing methods measure bandwidth consumption. However, consumption alone is insufficient to quantify bandwidth contention. Furthermore, existing methods diagnose bandwidth for the entire program execution, but lack the ability to associate bandwidth performance to the source code and data structures involved. To address these challenges, we propose DR-BW, a new tool based on machine learning to identify bandwidth contention in NUMA architectures and provide optimization guidance. DR-BW first trains a set of micro benchmarks and extracts useful features to identify bandwidth contention via a supervised machine learning model. Our experiments show that DR-BW achieves more than 96% accuracy. Second, DR-BW associates memory accesses that incur bandwidth contention with data objects, which provides intuitive guidance for optimization. Third, we apply DR-BW to a number of real benchmarks. Our optimization based on the insights obtained from DR-BW yields up to a 6.5× speedup in modern NUMA architectures.

ScrubJay: deriving knowledge from the disarray of HPC performance data

Modern HPC centers comprise clusters, storage, networks, power and cooling infrastructure, and more. Analyzing the efficiency of these complex facilities is a daunting task. Increasingly, facilities deploy sensors and monitoring tools, but with millions of instrumented components, analyzing collected data manually is intractable. Data from an HPC center comprises different formats, granularities, and semantics, and handwritten scripts no longer suffice to transform the data into a digestible form. We present ScrubJay, an intuitive, scalable framework for automatic analysis of disparate HPC data. ScrubJay decouples the task of specifying data relationships from the task of analyzing data. Domain experts can store reusable transformations that describe relations between domains. ScrubJay also automates performance analysis. Analysts provide a query over logical domains of interest, and ScrubJay automatically derives needed steps to transform raw measurements. ScrubJay makes large-scale analysis tractable, reproducible, and provides insights into HPC facilities.

A Flexible Data Model to Support Multi-domain Performance Analysis

Performance data can be complex and potentially high dimensional. Further, it can be collected in multiple, independent domains. For example, one can measure code segments, hardware components, data structures, or an application’s communication structure. Performance analysis and visualization tools require access to this data in an easy way and must be able to specify relationships and mappings between these domains in order to provide users with intuitive, actionable performance analysis results.

Curvature-Based Crease Surfaces for Wave Visualization

Author(s): Aldrich, G; Gimenez, A; Oskin, M; Strelitz, R; Woodring, J; Kellogg, LH; Hamann, B | Abstract: