Mohamed Helmy

Case Study: Spark GPU-Enabled Framework to Control COVID-19 Spread Using Cell-Phone Spatio-Temporal Data

Computers, Materials & Continua, 2020

Nowadays, the world is fighting a dangerous form of Coronavirus that represents an emerging pande... more Nowadays, the world is fighting a dangerous form of Coronavirus that represents an emerging pandemic. Since its early appearance in China Wuhan city, many countries undertook several strict regulations including lockdowns and social distancing measures. Unfortunately, these procedures have badly impacted the world economy. Detecting and isolating positive/probable virus infected cases using a tree tracking mechanism constitutes a backbone for containing and resisting such fast spreading disease. For helping this hard effort, this research presents an innovative case study based on big data processing techniques to build a complete tracking system able to identify the central areas of infected/suspected people, and the new suspected cases using health records integration with mobile stations spatio-temporal data logs. The main idea is to identify the positive cases historical movements by tracking their phone location for the last 14 days (i.e., the virus incubation period). Then, by acquiring the citizen's mobile phone locations for the same period, the system will be able to measure the Euclidean distances between positive case locations and other nearby people to identify the incontact suspected-cases using parallel clustering and classification techniques. Moreover, the daily change of the clusters size and its centroids will be used to predict new regions of infection, as well as, new cases. Moreover, this approach will support infection avoidance by alerting people approaching areas of high probability of infection using their mobile GPS location. This case study has been developed as a simulation system consisting of three components; positive cases/citizens movement's data generation subsystem, big data processing platform including CPU/GPU tasks, and data visualization/map geotagging subsystem. The processing of such a big data system requires intensive computing tasks. Therefore, GPU tasks carried out to achieve high performance and accelerate the data processing. According to the simulated system results, data partitioning and processing speed up measures have been examined.

Download

Comparative Study Load Balance Algorithms for Map Reduce Environment

International Journal of Computer Applications, 2014

MapReduce is a famous model for data-intensive parallel computing in shared-nothing clusters. One... more MapReduce is a famous model for data-intensive parallel computing in shared-nothing clusters. One of the main issues in MapReduce is the fact of depending its performance mainly on data distribution. MapReduce contains simple load balance technique based on FIFO job scheduler that serves the jobs in their submission order but unfortunately it is insufficient in real world cases as it missed many factors that impact the performance such as heterogeneity factor and data skewness, so Load balancing is important to make all resources utilized evenly and more efficiently. There are two main schemes in load balancing a-Static Load Balancing Schemes b-Dynamic load balancing. The main aim of this work is to study and compare existing Load Balance algorithms also to illustrate the features of Load Balance algorithms

Download

To appear in: Future Generation Computer Systems

Big Data represents a major challenge for the performance of the cloud computing storage systems.... more Big Data represents a major challenge for the performance of the cloud computing storage systems. Some distributed file systems (DFS) are widely used to store big data, such as Hadoop Distributed File System (HDFS), Google File System (GFS) 1 0 and others. These DFS replicate and store data as multiple copies to provide availability and reliability, but they increase storage and resources consumption.

Download

iHOME: Index-Based JOIN Query Optimization for Limited Big Data Storage

Query optimization in Big Data becomes a promising research direction due to the popularity of ma... more Query optimization in Big Data becomes a promising research direction due to the popularity of massive data analytical systems such as Hadoop system. The query optimization is getting hard to efficiently execute JOIN queries on top of Hadoop query language, Hive, over limited Big Data storages. According to our previous work, HiveQL Optimization for JOIN query over Multi-session Environment (HOME) system has been introduced over Hadoop system to improve its performance by storing the intermediate results to avoid repeated computations. Time overheads and Big Data storages limitation are considered the main drawback of the HOME system, especially in the case of using additional physical storages or renting extra virtualized storages. In this paper, an index-based system for reusing data called indexing HiveQL Optimization for JOIN R. Sahal ( ) · F. A. Omara over Multi-session Big Data Environment (iHOME) is proposed to overcome HOME overheads by storing only the indexes of the joined rows instead of storing the full intermediate results directly. Moreover, the proposed iHOME system addresses eight cases of JOIN queries which classified into three groups; Similar-to-iHOME, Compute-on-iHOME, and Filterof-iHOME. According to the experimental results of the iHOME system using TPC-H benchmark, it is found that the execution time of eight JOIN queries using iHOME on Hive has been reduced. Also, the stored data size in the iHOME system is reduced relative to the HOME system, as well as, the Big Data storage is saved. So, by increasing stored data size, the iHOME system guarantees the space scalability and overcomes the storage limitation.

Download

SOOM: Sort-Based Optimizer for Big Data Multi-Query

Mostly, sorting of data is a common operation in many applications, which causes the consumption ... more Mostly, sorting of data is a common operation in many applications, which causes the consumption of resources and thus leads to computation overheads. Regarding the context of Big Data multi-query, the shared sort operations are fairly large, which incur high-cost I/Os whether explicit or implicit. In particular, Big Data multiquery, including aggregation and sort operations, takes long execution time due to reshuffle of the same data multiple times using similar tasks. Therefore, exploiting the sharing data and the sharing sort opportunities of similar tasks can offer the possibility of reusing the previous results to optimize multi-query. For considering sharing data, our previous work, Multi-Query Optimization Using Tuple Size and Histogram (MOTH) system, has been introduced to consider the granularity of the sharing data opportunities among multi-query. However, time overheads regarding redundant data in-network movement (i.e., shuffling time to transfer intermediate data for sort operations) have not been considered. Therefore, the MOTH system has been extended to SOOM (Sort-Based Optimizer over MOTH) system to exploit sharing sort opportunities, including explicit sorts of sort queries and implicit sorts of aggregation queries. The proposed SOOM system consists of two additional modules to exploit sharing sort opportunities, namely query explorer and sort exploiter, which leverage our existing MOTH system to fulfill optimizing multiple aggregation and sort queries. The experimental evaluation has shown that the SOOM system outperforms the naive and the state-of-art techniques regarding query execution time among queries by 45% and 30%, respectively, while introducing maximal intermediate data size reduction by 67% and 61% in average, respectively, over Hadoop-like infrastructures.

Download

Exploiting coarse-grained reused-based opportunities in Big Data multi-query optimization

Multi-query optimization in Big Data becomes a promising research direction due to the popularity... more Multi-query optimization in Big Data becomes a promising research direction due to the popularity of massive data analytical systems (e.g., MapReduce and Flink). The multi-query is translated into jobs. These jobs are routinely submitted with similar tasks to the underling Big Data analytical systems. These similar tasks are considered complicated and computation overhead. Therefore, there are some existing techniques that have been proposed for exploiting sharing tasks in Big Data multi-query optimization (e.g., MRShare and Relaxed MRShare). These techniques are heavily tailored relaxed optimizing factors of fine-grained reused-based opportunities. In accordance with Big Data multi-query optimization, the existing fine-grained techniques are only concerned with equal tuples size and uniform data distribution. These issues are not applicable to the real-world distributed applications which depend on coarse-grained reused-based opportunities, such as non-equal tuples size and non-uniform data distribution. These two issues receive more-attention in Big Data multi-query optimization, to minimize the data read from or written back to Big Data infrastructures (e.g., Hadoop). In this paper, Multi-Query Optimization using Tuple Size and Histogram (MOTH) system has been proposed to consider the granularity of the reused-based opportunities. The proposed MOTH system exploits the coarse-grained of the fully and partially reused-based opportunities among queries with considering non-equal tuples size and non-uniform data distribution to avoid repeated computations. According to the proposed MOTH system, a combined technique has been introduced for estimating the coarse-grained reused-based opportunities horizontally and vertically. The horizontal estimation of non-equal tuples size has been done by extracting metadata in column-level, while the vertical estimation of non-uniform data distribution is concerned with using pre-computed histogram in row-level. In addition, the MOTH system estimates the coarse-grained reused-based opportunities with considering slow storage (i.e., limited physical resources or fewer allocated virtualized resources) to produce the accurate estimation of the reused results costs. Then, a cost-based heuristic algorithm has been introduced to select the best reused-based opportunity and generate an efficient multi-query execution plan. Because the partial reused-based opportunities have been considered, extra computations are needed to retrieve the non-derived results. Also, a partial reused-based optimizer has been tailored and added to the proposed MOTH system to reformulate the generated multi-query plan to improve the shared partial queries. According to the experimental results of the proposed MOTH system using TPC-H benchmark, it is found that multi-query execution time has been reduced by considering the granularity of the reused results.

Download

Case Study: Spark GPU-Enabled Framework to Control COVID-19 Spread Using Cell-Phone Spatio-Temporal Data

Nowadays, the world is fighting a dangerous form of Coronavirus that represents an emerging pande... more Nowadays, the world is fighting a dangerous form of Coronavirus that represents an emerging pandemic. Since its early appearance in China Wuhan city, many countries undertook several strict regulations including lockdowns and social distancing measures. Unfortunately, these procedures have badly impacted the world economy. Detecting and isolating positive/probable virus infected cases using a tree tracking mechanism constitutes a backbone for containing and resisting such fast spreading disease. For helping this hard effort, this research presents an innovative case study based on big data processing techniques to build a complete tracking system able to identify the central areas of infected/suspected people, and the new suspected cases using health records integration with mobile stations spatio-temporal data logs. The main idea is to identify the positive cases historical movements by tracking their phone location for the last 14 days (i.e., the virus incubation period). Then, by acquiring the citizen's mobile phone locations for the same period, the system will be able to measure the Euclidean distances between positive case locations and other nearby people to identify the incontact suspected-cases using parallel clustering and classification techniques. Moreover, the daily change of the clusters size and its centroids will be used to predict new regions of infection, as well as, new cases. Moreover, this approach will support infection avoidance by alerting people approaching areas of high probability of infection using their mobile GPS location. This case study has been developed as a simulation system consisting of three components; positive cases/citizens movement's data generation subsystem, big data processing platform including CPU/GPU tasks, and data visualization/map geotagging subsystem. The processing of such a big data system requires intensive computing tasks. Therefore, GPU tasks carried out to achieve high performance and accelerate the data processing. According to the simulated system results, data partitioning and processing speed up measures have been examined.

Download

Uploads

Papers by Mohamed Helmy

Log In