Cross-architecture performance predictions for scientific applications using parameterized models

Marin, Gabriel; Mellor-Crummey, John

doi:10.1145/1005686.1005691

Cited by 151 publications

(53 citation statements)

References 12 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Compared to the state-of-the-art techniques proposed in [9,28], our methodology is capable of reducing the prediction error by an order of magnitude on the instruction count, and it improves the accuracy of memory access pattern prediction by up to 1.3×. Additionally, in this work two case studies demonstrate how the methodology can analyze parallel applications.…”

Section: Introductionmentioning

confidence: 86%

“…Of particular interest are the values of p(d) when d is a power of two. These values can be used for inferring the miss rate for a cache size of d [28]. The probability p(d) is bounded between [0, 1].…”

Section: Predicting the Memory Access Patternmentioning

confidence: 99%

“…We use the artificial neural network ANN and the state-of-the-art methodology (SOTA) proposed in [28] as reference methodologies for comparison. In SOTA, different regression models are trained to predict p(d) over different regions of d. We focus our study on values of d that are powers of 2 as these values can be used for predicting the cache miss rate for caches having size d [20].…”

Section: Memory Access Patternmentioning

confidence: 99%

See 2 more Smart Citations

Scaling Properties of Parallel Applications to Exascale

Mariani

Anghel

Jongerius

et al. 2016

Int J Parallel Prog

View full text Add to dashboard Cite

A detailed profile of exascale applications helps to understand the computation, communication and memory requirements for exascale systems and provides the insight necessary for fine-tuning the computing architecture. Obtaining such a profile is challenging as exascale systems will process unprecedented amounts of data. Profiling applications at the target scale would require the exascale machine itself. In this work we propose a methodology to extrapolate the exascale profile from experimental observations over datasets feasible for today's machines. Extrapolation models are carefully selected by means of statistical techniques and a high-level complexity analysis is included in the selection process to speed up the learning phase and to improve the accuracy of the final model. We extrapolate run-time properties of the target applications including information about the instruction mix, memory access pattern, instruction-level parallelism, and communication requirements. Compared to state-of-the-art techniques, the proposed methodology reduces the prediction error by an order of magnitude on the instruction count and improves the accuracy by up to 1.3× for the memory access pattern, and by more than 2× for the communication requirements.

show abstract

Section: Introductionmentioning

confidence: 86%

Section: Predicting the Memory Access Patternmentioning

confidence: 99%

Section: Memory Access Patternmentioning

confidence: 99%

See 1 more Smart Citation

Scaling Properties of Parallel Applications to Exascale

Mariani

Anghel

Jongerius

et al. 2016

Int J Parallel Prog

View full text Add to dashboard Cite

show abstract

“…Several systems combine predictions of the sequential performance of single-node tasks performed by distributed cores with models of communication between them [6,28,46]. Similar cross-platform performance predictions for large-scale machines using a combination of known relative performance of the two systems and partial execution of the workload are described in [44].…”

Section: Related Workmentioning

confidence: 99%

“…Answering this question is very hard. Typical approaches include performing extensive performance evaluation or developing detailed models of applications [15,28], which are time-consuming, error-prone, and require detailed knowledge of each application and the machine it executes on.…”

Section: Introductionmentioning

confidence: 99%

Estima

Chatzopoulos

Dragojević

Guerraoui

2017

ACM Trans. Parallel Comput.

View full text Add to dashboard Cite

This article presents estima, an easy-to-use tool for extrapolating the scalability of in-memory applications. estima is designed to perform a simple yet important task: Given the performance of an application on a small machine with a handful of cores, estima extrapolates its scalability to a larger machine with more cores, while requiring minimum input from the user. The key idea underlying estima is the use of stalled cycles (e.g., cycles that the processor spends waiting for missed cache line fetches or busy locks). estima measures stalled cycles on a few cores and extrapolates them to more cores, estimating the amount of waiting in the system. estima can be effectively used to predict the scalability of in-memory applications for bigger execution machines. For instance, using measurements of memcached and SQLite on a desktop machine, we obtain accurate predictions of their scalability on a server. Our extensive evaluation shows the effectiveness of estima on a large number of in-memory benchmarks. INTRODUCTIONCommodity machines nowadays have hundreds of gigabytes of memory. This enables building performance-critical parallel applications, such as databases and key-value stores, that keep their datasets in main memory. This way, applications avoid slow secondary storage and networks, leaving the CPU as the main performance bottleneck [9,12,26,30]. Understanding the performance of these applications proves to be hard, since the number of CPU cores available during the deployment of a parallel application can be significantly higher than that during its development and testing. Applications developed today can be tested on machines with 16 or 24 cores, but in a few years the same applications are likely to be run on machines with 64 or even more cores.

show abstract

Predicting parallel application performance via machine learning approaches

Singh

Ïpek

McKee

et al. 2007

Concurrency and Computation

View full text Add to dashboard Cite

Consistently growing architectural complexity and machine scales make creating accurate performance models for large-scale applications increasingly challenging. Traditional analytic models are difficult and time-consuming to construct, and are often unable to capture full system and application complexity. To address these challenges, we automatically build models based on execution samples. We use multilayer neural networks, since they can represent arbitrary functions and handle noisy inputs robustly.In this thesis, we focus on two well known parallel applications whose variations in execution times are not well understood: SMG2000, a semicoarsening multigrid solver, and HPL, an open source implementation of LINPACK. We sparsely sample performance data on two radically different platforms across large, multi-dimensional parameter spaces and show that our models based on this data can predict performance within 2% to 7% of actual application runtimes.

show abstract

Cross-architecture performance predictions for scientific applications using parameterized models

Cited by 151 publications

References 12 publications

Scaling Properties of Parallel Applications to Exascale

Scaling Properties of Parallel Applications to Exascale

Estima

Predicting parallel application performance via machine learning approaches

Contact Info

Product

Resources

About