ORCID Profile
0000-0002-0844-5819
Current Organisation
Massey University
Does something not look right? The information on this page has been harvested from data sources that may not be up to date. We continue to work with information providers to improve coverage and quality. To report an issue, use the Feedback Form.
Publisher: Springer Science and Business Media LLC
Date: 12-2020
DOI: 10.1186/S40537-020-00388-5
Abstract: Big Data analytics for storing, processing, and analyzing large-scale datasets has become an essential tool for the industry. The advent of distributed computing frameworks such as Hadoop and Spark offers efficient solutions to analyze vast amounts of data. Due to the application programming interface (API) availability and its performance, Spark becomes very popular, even more popular than the MapReduce framework. Both these frameworks have more than 150 parameters, and the combination of these parameters has a massive impact on cluster performance. The default system parameters help the system administrator deploy their system applications without much effort, and they can measure their specific cluster performance with factory-set parameters. However, an open question remains: can new parameter selection improve cluster performance for large datasets? In this regard, this study investigates the most impacting parameters, under resource utilization, input splits, and shuffle, to compare the performance between Hadoop and Spark, using an implemented cluster in our laboratory. We used a trial-and-error approach for tuning these parameters based on a large number of experiments. In order to evaluate the frameworks of comparative analysis, we select two workloads: WordCount and TeraSort. The performance metrics are carried out based on three criteria: execution time, throughput, and speedup. Our experimental results revealed that both system performances heavily depends on input data size and correct parameter selection. The analysis of the results shows that Spark has better performance as compared to Hadoop when data sets are small, achieving up to two times speedup in WordCount workloads and up to 14 times in TeraSort workloads when default parameter values are reconfigured.
Publisher: Springer Science and Business Media LLC
Date: 14-08-2021
DOI: 10.1186/S40537-021-00499-7
Abstract: This article proposes a new parallel performance model for different workloads of Spark Big Data applications running on Hadoop clusters. The proposed model can predict the runtime for generic workloads as a function of the number of executors, without necessarily knowing how the algorithms were implemented. For a certain problem size, it is shown that a model based on serial boundaries for a 2D arrangement of executors can fit the empirical data for various workloads. The empirical data was obtained from a real Hadoop cluster, using Spark and HiBench. The workloads used in this work were included WordCount, SVM, Kmeans, PageRank and Graph (Nweight). A particular runtime pattern emerged when adding more executors to run a job. For some workloads, the runtime was longer with more executors added. This phenomenon is predicted with the new model of parallelisation. The resulting equation from the model explains certain performance patterns that do not fit Amdahl’s law predictions, nor Gustafson’s equation. The results show that the proposed model achieved the best fit with all workloads and most of the data sizes, using the R-squared metric for the accuracy of the fitting of empirical data. The proposed model has advantages over machine learning models due to its simplicity, requiring a smaller number of experiments to fit the data. This is very useful to practitioners in the area of Big Data because they can predict runtime of specific applications by analysing the logs. In this work, the model is limited to changes in the number of executors for a fixed problem size.
Publisher: MDPI AG
Date: 05-11-2021
DOI: 10.3390/BDCC5040065
Abstract: Big data frameworks play a vital role in storing, processing, and analysing large datasets. Apache Spark has been established as one of the most popular big data engines for its efficiency and reliability. However, one of the significant problems of the Spark system is performance prediction. Spark has more than 150 configurable parameters, and configuration of so many parameters is challenging task when determining the suitable parameters for the system. In this paper, we proposed two distinct parallelisation models for performance prediction. Our insight is that each node in a Hadoop cluster can communicate with identical nodes, and a certain function of the non-parallelisable runtime can be estimated accordingly. Both models use simple equations that allows us to predict the runtime when the size of the job and the number of executables are known. The proposed models were evaluated based on five HiBench workloads, Kmeans, PageRank, Graph (NWeight), SVM, and WordCount. The workload’s empirical data were fitted with one of the two models meeting the accuracy requirements. Finally, the experimental findings show that the model can be a handy and helpful tool for scheduling and planning system deployment.
Publisher: IEEE Comput. Soc
Date: 2002
No related grants have been discovered for Mohammad Rashid.