상세정보
Export to Refworks부가기능
Predicting Performance of Parallel Analytics and Irregular Computations
상세 프로파일
| 자료유형 | 학위논문 |
|---|---|
| 서명/저자사항 | Predicting Performance of Parallel Analytics and Irregular Computations. |
| 개인저자 | Zhu, Gangyi. |
| 단체저자명 | The Ohio State University. Computer Science and Engineering. |
| 발행사항 | [S.l.]: The Ohio State University., 2019. |
| 발행사항 | Ann Arbor: ProQuest Dissertations & Theses, 2019. |
| 형태사항 | 182 p. |
| 기본자료 저록 | Dissertations Abstracts International 81-06B. Dissertation Abstract International |
| ISBN | 9781392502648 |
| 학위논문주기 | Thesis (Ph.D.)--The Ohio State University, 2019. |
| 일반주기 |
Source: Dissertations Abstracts International, Volume: 81-06, Section: B.
Advisor: Agrawal, Gagan. |
| 이용제한사항 | This item must not be sold to any third party vendors. |
| 요약 | Scientific simulation and computation are becoming more complicated in recent years due to multiple reasons including new data processing paradigms, irregular computation pattern, evolving hardware development, and other factors. Since there is an increasingly large gap between the I/O performance and compute power so that the cost of writing and reading the vast amount of simulated data from disk is expensive, more attentions have been paid to in-situ analytics. Irregular computation, usually arising in domains that use sparse matrices, graphs, or other irregular data structures, introduces a variety of uncertainty to the performance of scientific computation. Moreover, the newly developed hardware platforms such as MIC and GPUs also impose challenges for understanding and improving performance of scientific applications. Hence, there is clearly a need for analyzing the impacts of these factors to scientific computations.We start from predicting performance of the disk-based and in-situ parallel data analytics implemented in MapReduce-like frameworks. We take two distinct approaches towards performance prediction. We first expand SKOPE (a SKeleton framewOrk for Performance Exploration) with performance models for disk data read, cache performance, and page fault penalty. Second, an analytical performance model is also developed.Next, we take irregular computations into consideration for performance prediction. Cache performance of irregular computations is highly input-dependent. Based on the sparse matrix view of irregular computation as well as the cache locality analysis, we propose a novel sampling approach named Adaptive Stratified Row sampling -- this method is capable of generating a representative sample that delivers cache performance similar to the original input. On top of our sampling method, we incorporate reuse distance analysis to accommodate different cache configurations with high efficiency. Besides, we modify SKOPE, a code skeleton framework, to predict the execution time for irregular applications with the predicted cache performance.We extend the work of modeling irregular computations to the SIMD scenario. Our first insight is that developing a universal sampling approach for all sparse matrices is unpractical. According to the non-zero distribution of the sparse matrix, we propose two novel sampling strategies: Stride Average sampling and Random Tile sampling, which are suitable for uniform and skewed sparse matrices respectively. To help categorize a sparse matrix as uniform or skewed, we introduce clustering coefficient as an important feature which can be propagated into the decision tree model. We also adapt Random Node Neighbor sampling approach for efficient estimation of clustering coefficient.Finally, we target another topic of irregular computation on GPUs: sparse matrix format selection for Sparse Matrix-Vector Multiplication (SpMV). Based on the storage properties and processing granularity of different formats, we develop three novel sampling schemes: Row Crop sampling, Random Warp sampling, and Diagonal Align sampling. Then we obtain the base performance for each format by executing SpMV over the generated samples. The best format is predicted by scaling the base performance based on the difference of parallelism between the original matrix and the sampled one. |
| 일반주제명 | Computer engineering. Computer science. |
| 언어 | 영어 |
| 바로가기 | 
: 이 자료의 원문은 한국교육학술정보원에서 제공합니다. |
