Euro-Par 2011 Parallel Processing

Name: Euro-Par 2011 Parallel Processing | 17th International Euro-ParConference, Bordeaux, France, August 29 - September 2, 2011, Proceedings, Part I
Brand: Springer
Price: 53.49 EUR
Availability: OnlineOnly

17th International Euro-ParConference, Bordeaux, France, August 29 - September 2, 2011, Proceedings, Part I

Emmanuel Jeannot Raymond Namyst Jean Roman(Herausgeber*in)

Springer (Verlag)

Erschienen am 12. August 2011

XXXIII, 598 Seiten

E-Book

PDF mit Wasserzeichen-DRM

Systemvoraussetzungen

978-3-642-23400-2 (ISBN)

53,49 €inkl. 7% MwSt.

Systemvoraussetzungen

für PDF mit Wasserzeichen-DRM

E-Book Einzellizenz

Als Download verfügbar

Beschreibung

Weitere Details

Weitere Ausgaben

Inhalt

Title
Preface
Organization
Table of Contents
Topic 1: Support Tools and Environments
Introduction
Run-Time Automatic Performance Tuning for Multicore Applications
Introduction
The Perpetuum Run-Time Application Tuner
Preparing Applications for Online Tuning
Perpetuum in Action: Automated Online-Tuning in Parallel Compression
Environment
Scenario 1: Tuning a Single Process
Scenario 2: Simultaneously Auto-tuning Two Processes
Scenario 3: Simultaneously Auto-tuning Two Processes Starting with a Time Lag
Summary
Automated Online-Tuning in Parallel Video Processing
Scenario 1: Tuning a Single Process
Scenario 2: Simultaneously Tuning Two Processes Starting with a Time Lag
Related Work
Conclusion
References
Exploiting Cache Traffic Monitoring for Run-Time Race Detection
Introduction
Assumptions and Requirements
Software
Hardware
Monitoring Cache Traffic to Detect and Heal Races
Race Detection
Race Healing
TheDetector
Evaluation
Setup
Results
Related Work
Conclusion
References
Accelerating Data Race Detection with Minimal Hardware Support
Introduction
Background
Minimal Hardware Support for Data Race Detection
AccessedBefore (AccB) Algorithm
Sources of Inaccuracy
Implementation
Hardware Support
Software Layer
Optimizations
System Issues
Experimental Setup
Evaluation
AccB versus HapB
Overheads Characterization
Accuracy Characterization
Related Work
Conclusions
References
Quantifying the Potential Task-Based Dataflow Parallelism in MPI Applications
Introduction
SMPSs Programming Model
Motivation
Framework
Input Code
Code Translator
Tracer
Replay Simulator
Experiments
Results
Related Work
Conclusion
References
Event Log Mining Tool for Large Scale HPC Systems
Introduction
Related Work
Methodology
Offline Clustering
Splitting Process
Output
Online Clustering
Log Files
Results
Offline
Online
Conclusion and Future Work
References
Reducing the Overhead of Direct Application Instrumentation Using Prior Static Analysis
Introduction
Related Work
A Configurable Instrumenter
Adapter Specification
Filter Specification
Filter Criteria
Evaluation
Conclusion and Future Work
References
Topic 2: Performance Prediction and Evaluation
Introduction
Reducing Energy Usage with Memory and Computation-Aware Dynamic Frequency Scaling
Introduction
Methodology
Benchmarking for Power and Performance
Application Characterization
Experimental Results
Drawing Conclusions about System Behavior
Energy-Optimal Clock Frequency Selection
Technique Validation
Related Work
Future Work
Conclusions
References
A Contention-Aware Performance Model for HPC-Based Networks: A Case Study of the InfiniBand Network
Introduction
Background
Elements Influencing Network Contention
Methodology
Dynamic Contention Graph
Sequence of Linear Models
Approximation of Penalty Coefficients
Modelling Penalty Coefficients over InfiniBand
InfiniBand Network Testbed
Penalty Coefficients and Model for InfiniBand
Examples and Validation
Conclusion and Future Work
References
Using the Last-Mile Model as a Distributed Scheme for Available Bandwidth Prediction
Introduction
Related Works
Latency Estimation
Bandwidth Estimation
Last-Mile Bandwidth Prediction Model
Last-Mile Model
Initial Values
Iterative Procedure
Evaluation
Methodology
Parameter Tuning
Comparison Methods
Evaluation Results
Concluding Remarks
References
Self-stabilization versus Robust Self-stabilization for Clustering in Ad-Hoc Network
Introduction
Overview of the Studied Clustering Protocols
Model and Simulation Remarks
Observed Metrics
Simulation Results and Performances Analysis
Average Number of Cluster-Heads
Availability of Minimum Service
Availability of Optimum Service
Concluding Remarks
References
Multilayer Cache Partitioning for Multiprogram Workloads
Introduction
Motivational Example for Multilayer Partitioning
Dynamic Performance Model
Proposed Partitioning Algorithm
Experimental Evaluation
Implementation and Setup
Results
Concluding Remarks
References
Backfilling with Guarantees Granted upon Job Submission
Introduction
Algorithms
Prioritized Compression
Delayed Compression
Related Work
Experimental Results
Increasing Responsiveness
Favoring Wide Jobs
Scheduler Running Time
Discussion
References
Topic 3: Scheduling and Load Balancing
Introduction
Greedy "Exploitation" Is Close to Optimal on Node-Heterogeneous Clusters
Introduction
Formal Details
Work Production under the LIFO and FIFO Protocols
The LIFO Protocol Is Approximately Optimal
A Lower Bound on the LIFO Work Production W(L)(C
L)
An Upper Bound on the FIFO Work Production W(F)(C
L)
The LIFO-FIFO Bounding Ratio
Conclusions
References
Scheduling JavaSymphony Applications on Many-Core Parallel Computers
Introduction
Related Work
JavaSymphony
JavaSymphony Scheduler
System Architecture
Scheduling Methodology
Algorithm
Experiments
Experimental Methodology
Communication-Intensive Applications
Training Experiments.
Validation Experiments.
Computation-Intensive Applications
Training Experiments.
Validation Experiments.
Conclusions
Assessing the Computational Benefits of AREA-Oriented DAG-Scheduling
Introduction
Background
Finding Good AO-Schedules Efficiently
Experiments to Assess the Quality of aoh
Experimental Design
Experimental Methodology
Experimental Results and Discussion
Conclusion
Analysis and Modeling of Social Influence in High Performance Computing Workloads
Introduction
Data Sources
Social Influence Model
Analysis of Social Influence
Community Extraction from HPC Workloads
Power-Law Distribution of Discovered Communities
Design of Online Learning Mechanism
Related Work
Conclusions and Future Work
Work Stealing for Multi-core HPC Clusters
Introduction
Work Stealing
Design for Our Approach
Shared Memory Design
Distributed Memory Design
Combined Approach
Evaluation
UTS
Results
Related Work
Conclusion and Future Work
A Dynamic Power-Aware Partitioner with Task Migration for Multicore Embedded Systems
Introduction
Related Work
System Model
Task Real-Time Behavior
Power-Aware Scheduler
Partitioning Heuristics with Task Migration
Extending Worst Fit to Support Task Migration
Dynamic Partitioner
Experimental Results
Impact of Applying Migrations at Different Points of Time
Comparing DP versus WF Variants
Conclusions
Exploiting Thread-Data Affinity in OpenMP with Data Access Patterns
The Data Access Pattern Approach
Data Access Pattern Definition
Runtime Extensions to Exploit Patterns
Iteration Space Partitioning
A Dynamic Scheduling Policy for Pattern Enabled OpenMP Runtimes
Work Stealing Strategy
Experimental Results
Benchmark Suite
Performance Analysis
Remote Memory Access Analysis
Related Work
Conclusions
Workload Balancing and Throughput Optimization for Heterogeneous Systems Subject to Failures
Introduction
Framework and Optimization Problems
Complexity Results
Complexity of the MinPer (*,fi,*) Problems
Complexity of the MinPer (*,fi,u,*) Problems
Heuristics and Simulations
Polynomial Time Heuristics
Simulations
Conclusion
On the Utility of DVFS for Power-Aware Job Placement in Clusters
Introduction
Related Work
Power-Aware Job Placement with DVFS
Problem Statement
Problem Formulation
DVFS/DFS Model
Numerical Results
Experimental Methodology
Results for Small Instances
Results for Large Instances
Conclusion
Topic 4: High-Performance Architecture and Compilers
Introduction
Filtering Directory Lookups in CMPs with Write-Through Caches
Introduction
Motivation
Filtering Mechanism
Overview
Filter Operation
Filter States
Filter Overhead
Evaluation
Chip Multiprocessor Model
Methodology
Filter Coverage
Performance
Power Consumption
Related Work
Conclusions
References
FELI: HW/SW Support for On-Chip Distributed Shared Memory in Multicores
Introduction
On-Chip Distributed Shared Memory
Chip MultiProcessor Architectures
Single Global Address Space On-Chip
FELI - Operating System Support for Locality Management
L0 Cache
Discussion on DSM Architecture Parameters
Methodology
Experimental Evaluation
Related Work
Conclusions
References
Token3D: Reducing Temperature in 3D Die-Stacked CMPs through Cycle-Level Power Control Mechanisms
Introduction
Background and Related Work
Power and Thermal Control in Microprocessors
Building a 3D Die-Stacked Processor
3D Integration Technology
Thermal Control in 3D Die-Stacked Processors
Token3D: Balancing Temperature on 3D-Staked Designs
Token3D Implementation Details
Experimental Results
Simulation Environment
Effects of Token3D on Peak Temperature
Further Temperature Optimizations
Conclusions
References
Bandwidth Constrained Coordinated HW/SW Prefetching for Multicores
Introduction
Background and Methodology
Prefetching
Experimental Setup
Empirical Motivation
Prefetching Benefits
Off-Chip Bandwidth Effects
Prefetch Request Priority
Bandwidth Aware Prefetching
Core-Level Prefetch Manager
Prefetch Levels
Global Prefetch Manager
Experimental Evaluation
Related Work
Concluding Remarks
References
Unified Locality-Sensitive Signatures for Transactional Memory
Introduction
Background and Related Work
Unified Signature Design
Hardware Evaluation
False Positive Analysis
Evaluation
Methodology
Unified Signature Results
Unified Locality-Sensitive Signature Results
Conclusions
References
Using Runtime Activity to Dynamically Filter Out Inefficient Data Prefetches
Introduction
Related Works
Correlation between Runtime Activity and Prefetch Efficiency
Adaptive Prefetching Method Based on Runtime Activity
Experimental Evaluation
Experimental Environment
Experimental Results
Conclusion
References
Topic 5: Parallel and Distributed Data Management
Introduction
Distributed Scalable Collaborative Filtering Algorithm
Introduction
Related Work
Background and Notation
Optimized Distributed Co-clustering Algorithm
Parallel Time Complexity Analysis
Optimum Thread Distribution
Results and Analysis
Strong Scalability
Weak Scalability
Data Scalability
Conclusions and Future Work
References
Compressing the Incompressible with ISABELA: In-situ Reduction of Spatio-temporal Data
Introduction
A Motivating Example
Problem Statement
Theory and Methodology
Sorting-Based Data Transformation
Cubic B-Splines Fitting
Maximizing Compression Ratio via Window Splitting
Error Quantization for Guaranteed Point-by-Point Accuracy
Exploiting -Encoding for Temporal Index Compression
Results
Per Window Accuracy
Trade-Off between Compression and Per Point Accuracy
Effect of -encoding on Index Compression
Compression Time
Performance for Fixed Compression
Related Work
Summary
References
kNN Query Processing in Metric Spaces Using GPUs
Introduction
Similarity Search Background and Related Work
List of Clusters (LC)
Sparse Spatial Selection (SSS-Index)
Graphic Processing Units (GPU)
GPU Mapping of k-Nearest Neighbor Algorithms
Exhaustive Search Algorithm
LC
SSS-Index
Experimental Results
Conclusions
References
An Evaluation of Fault-Tolerant Query Processing for Web Search Engines
Introduction
Indexing and Ranking
Experimental Framework
Process-Oriented Discrete-Event Simulator
Simulating Failures
Simulator Validation
Comparative Evaluation
Concluding Remarks
References
Topic 6: Grid Cluster and Cloud Computing
Introduction
Self-economy in Cloud Data Centers: Statistical Assignment and Migration of Virtual Machines
Introduction
Assignment and Migration of Virtual Machines
Assignment Procedure
Migration Procedure
Performance Evaluation
Related Work
Conclusion and Future Work
References
An Adaptive Load Balancing Algorithm with Use of Cellular Automata for Computational Grid Systems
Introduction
Cellular Automata
The Proposed Load Balancing Algorithm
Simulation and Results
Simulation Model
Conclusion
References
Shrinker: Improving Live Migration of Virtual Clusters over WANs with Distributed Data Deduplication and Content-Based Addressing
Introduction
Background and Related Work
Architecture of Shrinker
Architecture Overview
Security Considerations
Implementation and Performance Evaluation
Implementation
Evaluation Methodology
Performance Results
Conclusion
References
Maximum Migration Time Guarantees in Dynamic Server Consolidation for Virtualized Data Centers
Introduction
Related Work
Server Consolidation Algorithm
First Phase: Minimizing Migration Time
Second Phase: Minimizing the Number of Physical Servers
Evaluation
Conclusion and Future Work
References
Enacting SLAs in Clouds Using Rules
Introduction
Related Work
Escalation Levels
Rule-Based Approach for VM Level
Prerequisites
Design and Implementation
Evaluation
Utility-Driven Evaluation
Performance-Driven Evaluation
Conclusion and Outlook
References
DEVA: Distributed Ensembles of Virtual Appliances in the Cloud
Introduction
System Overview
DEVA Manager
Mapping of DEVAs
DEVAs across Heterogeneous Resources
DEVA Agents
Experimental Results
Overhead Measurement
Isolation and QoS Conservation
Use of Heterogeneous DEVAs and Resources
Related Work
Conclusions and Future Work
References
Benchmarking Grid Information Systems
Introduction
Related Work
Methodology
Benchmarking MDS and BDII
Background
Experiment Setup
Query Response Time
Quality of Information
Discussion
Conclusion
References
Green Cloud Framework for Improving Carbon Efficiency of Clouds
Introduction
Related Work
Carbon Aware Green Cloud Architecture
Third Party: Green Offer Directory and Carbon Emission Directory
User: Green Broker
Provider: Green Middleware
Case Study: IaaS Cloud
Carbon Efficient Green Policy (CEGP)
Performance Evaluation and Results
Comparison of CEGP with Performance-Based Algorithm (EST)
Effect of Relationship between CO2 Emission Rate and Datacenter Power Efficiency DCiE
Conclusion
References
Optimizing Multi-deployment on Clouds by Means of Self-adaptive Prefetching
Introduction
Our Approach
Design Principles
Architecture
Implementation
Experimental Evaluation
Experimental Setup
Performance of Multi-deployment
Related Work
Conclusions
References
Topic 7: Peer to Peer Computing
Introduction
Combining Mobile and Cloud Storage forProviding Ubiquitous Data Access
Introduction
System Design
Architecture
Synchronization Process
Advanced Mechanism for Using Mobile Phones
Minimizing Communications
Improving Storage Usage
Evaluation
Importance of DSV and DTC
Performance Impact of the DSV and DTC
Related Work
Final Remarks
References
Asynchronous Peer-to-Peer Data Mining with Stochastic Gradient Descent
Introduction
System and Data Model
Background
Related Work
The Algorithm
Experimental Results
Scenarios
Metrics
Results
Conclusions
References
Evaluation of P2P Systems under Different Churn Models: Why We Should Bother
Introduction
Related Work
Models and Distributions
Churn Models
ON/OFF Distributions
Comparative Analysis
Poissonity in Arrivals
Availability Inter-dependence: A First Difference
Reliability: A Second Difference
Conclusions
References
Topic 8: Distributed Systems and Algorithms
Introduction
Productive Cluster Programming with OmpSs
Introduction
OmpSs: From Multicores to Clusters
Overview
Example
Implementation
Evaluation
Methodology
Results
Related Work
Conclusions and Future Work
References
On the Use of Cluster-Based Partial Message Logging to Improve Fault Tolerance for MPI HPC Applications
Introduction
Context
Communication Patterns
Partitioning for Partial Message Logging Protocols
Two Possible Approaches
Bisection-Based Partitioning
Evaluation
Defining a Cost Function
Results
Conclusion
References
Object Placement for Cooperative Caches with Bandwidth Constraints
Introduction
System Model and Problem Formulation
Hardness Result and Approximation Algorithm
Hardness Proof
Efficient Placement Algorithm
Approximation Ratio
Extensions to Algorithm PA
Empirical Evaluation
Conclusion
References
Author Index

Systemvoraussetzungen

Als PDF speichern Als Link merken