Computer Science Focus on Research - University of Houston
Skip to main content

Computer Science Focus on Research

When: Wednesday, February 20, 2019
Where: PGH 563
Time: 11:00 AM


RRP Edge Computing System: Architecture and challenges

Guangli Da, PhD Student

Due to the popularity of the Internet of Things, more and more computing resources are needed for smart devices. To relieve the dependency on clouds and to lower the cost of equipment, we propose a Regularity-based Resource Partition Edge Computing System (RRP-ECS). In this system, RRP divides each CPU into 2 partitions, one of which handles the basic control, while the other functions as a core of a virtual machine dealing with intensive tasks. We introduce a centralized architecture and briefly discusses the challenges RRP-ECS is faced with in the task scheduling problem.

Guangli Dai received his B.S. degree in Computer Science from Southeast University, Nanjing, China, in 2017. He is currently pursuing the Ph.D. degree in Computer Science in University of Houston, USA. He is now a member of Real-Time Systems Laboratory in University of Houston. His research interests are in the areas of energy-efficient scheduling, real-time scheduling, virtualized systems and embedded systems.

Overlapped Two-Phase algorithm: Improving MPI-IO collective write operation performance

Raafat Feki, PhD Student

Massively parallel computers are increasingly used nowadays to solve large problems. The amount of data resulted from this computational power is exponentially growing and requires an important amount of time to be written/read into/from drive disks. The limited I/O capacities has become a very critical issue even with parallelism and hence we need more optimization techniques to be applied to the existing Parallel I/O algorithms. Within this context, the OpenMPI implementation proposed a class of optimizations called collective I/O. The most used algorithm illustrating this technique is Two-Phase I/O. It consist of a shuffle phase where compute nodes exchange the data over the network and an access phase, where the I/O operations are issued to the file system. The main drawback of the two-phase algorithm is the communication cost. Therefore, we suggested to overlap these two phases so we can reduce the overhead caused by the data transfer between the processors. As a first step, we implemented the overlapping technique using different approaches. Then, we evaluated each possible implementation using existing parallel I/O benchmarks in order to extract the best version.

Raafat Feki received his Engineering degree in computer science from the National School of Computer Sciences, Tunisia in 2015. Subsequently, he worked for over a year on decoding data received from real-time data feed connectivity to exchanges at FIS Tunis. Raafat is currently a Ph.D. student and Research Assistant at Department of Computer Science at University of Houston. His research interests include high performance computing, parallel I/O and OpenMPI development.

Efficient Distributed Community Detection in the Stochastic Block Model

Reza Fathi, PhD Student

Designing effective algorithms for community detection is an important and challenging problem in large-scale graphs, studied extensively in the literature. Various solutions have been proposed, but many of them are centralized with expensive procedures (requiring full knowledge of the input graph) and have a large running time. We present a distributed algorithm for community detection in the stochastic block model (also called {\em planted partition model}), a widely-studied and canonical random graph model for community detection and clustering. Our algorithm is based on random walks, and is localized and lightweight, and easy to implement.

Reza Fathi is a Ph.D. student working with Dr. Gopal Pandurangan. His research interests include Algorithm Design, Big Data Analytics, and Computer Security. His current research is focused on the design of distributed graph algorithms.

Light-Weight DBMS for Network Monitoring

Steve Aigbe, PhD Student

Monitoring networks is necessary to determine time periods of high usage, to identify devices consuming high bandwidth or uncovering security issues. In general, this is done with OS and networking utilities, which work directly on network packets or log files. Unfortunately, the spectrum of potential analyses is limited because the volume of data is significant and there is some programming involved. We present an alternative system that can monitor a local network evaluating SQL queries on a moving time window. Our system exploits a lightweight DBMS, which has low resource (CPU, RAM) consumption and provides basic SQL queries. Our system features important optimizations and recommended parameter settings to provide guaranteed performance. We show our system is capable of monitoring a small to medium-sized network with a set of complementary queries, running on a small computing device (i.e. a minimal, bare-bones computer). Our system can refresh the database with new network data and refresh visualization in a GUI in 10-30 seconds.

Steve Aigbe studied at the University of Ibadan, Nigeria where he received an undergraduate degree in Computer Science. Thereafter, he worked as an Integration Engineer both at Ericsson and Mobile Technologies Co. Ltd. From the fall of 2018, he started to pursue his PhD degree at the Department of Computer Science at the University of Houston. His current research is on analytics on streams in the DBMS Research Lab.