When Parallel Speedups Hit the Memory Wall

Furtunato, Alex F. A.; Georgiou, Kyriakos; Eder, Kerstin; Xavier‐de‐Souza, Samuel

doi:10.1109/access.2020.2990418

Cited by 7 publications

(4 citation statements)

References 31 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…In future work, we intend to evolve this tool to include the ability to predict speedup and efficiency from a few samples using state-of-the-art prediction models present in the literature [26,27]. The idea is to present the general behavior of the program and its scalability trend and reduce the execution time necessary to compose a comprehensive analysis.…”

Section: Discussionmentioning

confidence: 99%

A Minimally Intrusive Approach for Automatic Assessment of Parallel Performance Scalability of Shared-Memory HPC Applications

Silva

Valderrama

et al. 2022

Electronics

Self Cite

View full text Add to dashboard Cite

High-performance computing systems have become increasingly dynamic, complex, and unpredictable. To help build software that uses full-system capabilities, performance measurement and analysis tools exploit extensive execution analysis focusing on single-run results. Despite being effective in identifying performance hotspots and bottlenecks, these tools are not sufficiently suitable to evaluate the overall scalability trends of parallel applications. Either they lack the support for combining data from multiple runs or collect excessive data, causing unnecessary overhead. In this work, we present a tool for automatically measuring and comparing several executions of a parallel application according to various scenarios characterized by the input arrangements, the number of threads, number of cores, and frequencies. Unlike other existing performance analysis tools, the proposed work covers some gaps in specialized features necessary to better understand computational resources scalability trends across configurations. In order to improve scalability analysis and productivity over the vast spectrum of possible configurations, the proposed tool features automatic instrumentation, direct mapping of parallel regions, accuracy-preserving data reductions, and ease of use. As it aims at accurately understanding scalability trends of parallel applications, detailed single-run performance analyses show minimal intrusion (less than 1% overhead).

show abstract

Section: Discussionmentioning

confidence: 99%

A Minimally Intrusive Approach for Automatic Assessment of Parallel Performance Scalability of Shared-Memory HPC Applications

Silva

Valderrama

et al. 2022

Electronics

Self Cite

View full text Add to dashboard Cite

show abstract

“…It is the only parameter that characterizes the application. Indeed, by modelling complex parallel applications using their parallel fraction only, we are neglecting the effects that a frequency change has on the performance of the memory hierarchy, on the parallel overhead, and on the distribution of load across the heterogeneous cores [36]. Besides, as those features should limit the parallel speedup, it is expected that the sequential fraction includes them.…”

Section: Application Performance Modellingmentioning

confidence: 99%

Performance and Energy Trade-Offs for Parallel Applications on Heterogeneous Multi-Processing Systems

et al. 2020

Self Cite

View full text Add to dashboard Cite

This work proposes a methodology to find performance and energy trade-offs for parallel applications running on Heterogeneous Multi-Processing systems with a single instruction-set architecture. These offer flexibility in the form of different core types and voltage and frequency pairings, defining a vast design space to explore. Therefore, for a given application, choosing a configuration that optimizes the performance and energy consumption is not straightforward. Our method proposes novel analytical models for performance and power consumption whose parameters can be fitted using only a few strategically sampled offline measurements. These models are then used to estimate an application’s performance and energy consumption for the whole configuration space. In turn, these offline predictions define the choice of estimated Pareto-optimal configurations of the model, which are used to inform the selection of the configuration that the application should be executed on. The methodology was validated on an ODROID-XU3 board for eight programs from the PARSEC Benchmark, Phoronix Test Suite and Rodinia applications. The generated Pareto-optimal configuration space represented a 99% reduction of the universe of all available configurations. Energy savings of up to 59.77%, 61.38% and 17.7% were observed when compared to the performance, ondemand and powersave Linux governors, respectively, with higher or similar performance.

show abstract

“…Variations of aspects such as memory size and hierarchy, number of cores, and input size may have very diverse effects in parallel software performance [22]. Since the relation between the chunk size of the parallel loops and the total execution time of a program is unknown, using a stochastic optimization method to find an optimal chunk size is an alternative.…”

Section: Csa-based Auto-tuningmentioning

confidence: 99%

Auto-Tuning of Dynamic Scheduling Applied to 3D Reverse Time Migration on Multicore Systems

et al. 2020

Self Cite

View full text Add to dashboard Cite

Reverse time migration (RTM) is an algorithm widely used in the oil and gas industry to process seismic data. It is a computationally intensive task that suits well in parallel computers. Methods such as RTM can be parallelized in shared memory systems through scheduling iterations of parallel loops to threads. However, several aspects, such as memory size and hierarchy, number of cores, and input size, make optimal scheduling very challenging. In this paper, we introduce a run-time strategy to automatically tune the dynamic scheduling of parallel loops iterations in iterative applications, such as the RTM, in multicore systems. The proposed method aims to reduce the execution time of such applications. To find the optimal granularity, we propose a coupled simulated annealing (CSA) based auto-tuning strategy that adjusts the chunk size of work that OpenMP parallel loops assign dynamically to worker threads during the initialization of a 3D RTM application. Experiments performed with different computational systems and input sizes show that the proposed method is consistently better than the default OpenMP schedulers, static, auto, and guided, causing the application to be up to 33% faster. We show that the possible reason for this performance is the reduction of cache misses, mainly level L3, and low overhead, inferior to 2%. Having shown to be robust and scalable for the 3D RTM, the proposed method could also improve the performance of similar wave-based algorithms, such as full-waveform inversion (FWI) and other iterative applications.

show abstract

When Parallel Speedups Hit the Memory Wall

Cited by 7 publications

References 31 publications

A Minimally Intrusive Approach for Automatic Assessment of Parallel Performance Scalability of Shared-Memory HPC Applications

A Minimally Intrusive Approach for Automatic Assessment of Parallel Performance Scalability of Shared-Memory HPC Applications

Performance and Energy Trade-Offs for Parallel Applications on Heterogeneous Multi-Processing Systems

Auto-Tuning of Dynamic Scheduling Applied to 3D Reverse Time Migration on Multicore Systems

Contact Info

Product

Resources

About