Research

I am an Assistant Professor in the Department of Math and Computer Science at Athens State University. This page is intended to give the reader an idea about some of the research topics that I’m investigating at the moment.

Professional activities

Selected publications

Courses taught

  • Athens State University
    • CS305: Concepts of Computer Programming: First course overview of computer science for new CS majors
    • CS317: Computer Science I: Introduction to computer science for computer science and mathematics majors.
    • CS417: Object-oriented Design and Programming: A study of the issues involved in object-oriented analysis, design, and programming.
    • CS451: Software Engineering: Introduction to large scale software development and the management issues involved with such projects.
    • CS452: Senior Software Project: A capstone project course for computer science majors.
  • University of Louisiana at Lafayette
    • CMPS 301: Computing for the Natural Sciences: Introduction to computer science for engineering and science majors.
    • CMPS 352: Introduction to Scientific Computing: Introduction to computational science for computer science majors focusing on numerical analysis and scientific visualization.
    • CMPS 353: Principles of File Organization and Processing: File structures – their
      manipulation and management, application to commercial systems, techniques for data storage and
      retrival.

Current topics of interest

Server Power Management

It is difficult in practice to achieve efficient power management as data centers usually over-provision their power capacity to address the worst case scenarios. This results in either waste of considerable power budget or severe under-utilization of capacity. Thus, it is critical to quantitatively understand the relationship between power consumption and thermal envelope at the system level so as to optimize the use of deployed power capacity in the data center. Chip multiprocessors (CMP) and processors employing simultaneous multi-threading (SMT) have become common in server blades due to advantages these devices have in low design cost and better performance. However, the greater chip complexity entailed by many-core processors lead to larger power envelopes, elevated peak chip die temperatures, and imbalanced thermal gradients. So, as the workload increases on the server, so does the thermal stress placed on the processor with the resulting probability of damage to the machine.

Modern processors crudely manage this problem through Dynamic Thermal Management (DTM) where the processor monitors the die temperature and dynamically adjusts the processor voltage and frequency (DVFS) to slow down the processor. However, DVFS has a significant negative impact on application performance and system reliability. Thus, pro-active scheduling techniques that avoid thermal emergencies are preferable to reactive hardware techniques such as DTM. The component of the operating system most aware of the existence of multiple cores is the thread scheduler.

In our work, we introduce two techniques for thermal management that minimizes server energy consumption by (1) for each logical CPU, selecting the next thread to execute based upon an estimate of which thread has the least probability of causing a DTM in the next quantum, and (2) adjusting the load balance allocation of threads to available logical CPUs so as to migrate workload away from thermally overextended resources on the processor. Intelligent thermally-aware scheduling decisions requires an full-system model of energy consumption based on computational load of system that can be used to effectively predict future energy consumption and resulting changes in thermal load. It relates server energy consumption to its overall thermal envelope, establishing an energy relationship between the workload and the overall thermodynamics of the system. Our analysis of experimental measurements of key processor performance counter readings and performance metrics reveals that the measured readings and metrics do not possess linearity and are chaotic in nature.

Thus, our development of a Chaotic Attractor Predictor (CAP) that takes into account key thermal indicators (like ambient temperatures and die temperatures) and system performance metrics (like performance counters)
for system energy consumption estimation within a given power and thermal envelope. It captures the chaotic dynamics of a server system without complex and time-consuming corrective steps usually required by
linear prediction to account for the effects of non-linear and chaotic behavior in the system, exhibiting polynomial time complexity. Effective scheduling can result from taking advantage of the proposed CAP when
dispatching jobs to confine server power consumption within a given power budget and thermal envelope while minimizing impact upon server performance.

Wireless Sensor Networks

I’ve been spending some time recently working on a project at CACS that’s looking at applications of wireless networking to petroleum engineering. One aspect of this work is looking at applications of wireless sensor networks to platform environmental monitoring. You can see what we’re doing at UcOMS.org.

I’m looking into research problems related to routing, time synchronization, and power management of nodes in the network. You can see some of the work I’ve been doing in this area by looking at the publications list here on this page.

Technology in Retailing

Take a look around your local grocery store. Most people don’t really notice just how retailing has become technology driven. How people shop is changing because of new technologies like self-scanning and customer loyalty marketing system. Beyond the front of the store, the widespread of data warehouses and customer relationship management packages has allowed the retailer to severely optimize the supply chain.

Interesting Links