CSE Web Menu mobileicon

Professors Khan and Gokhale Receive Air Force Office of Scientific Research Grant

Congratulations to Professor Mohammad Maifi Khan and Professor Swapna Gokhale for receiving an Air Force Office of Scientific Research grant entitled: Performance Analysis and Diagnosis of Cloud-based DDDAS Applications. This project focuses on optimal resource allocation, tuning and troubleshooting of system performance in cloud settings, especially for DDDAS applications, which can be challenging for several reasons. For instance, in DDDAS applications, the sampling rate of different sensors and sensing modalities may change in response to the criticality of the situation or due to changes in the communication bandwidth and/or sensor utility function, which may cause a cascading effect on the cloud side, and significantly affect the systems overall performance. Also, at the application layer, different data processing algorithms with dramatically different execution times (e.g., motif mining vs. image analysis) can impact the load distribution on the servers. Moreover, growing technological trends like virtualization, wide adoption of parallel and multi-threaded programs due to the advent of multicore technologies, and increasingly larger scales are making it extremely hard to ascertain and troubleshoot suboptimal performance problems, especially in cloud settings. To address this growing challenge, this project investigates automated scalable solutions for performance troubleshooting that will not only identify the root cause but also answer questions such as “How is the changed sampling rate affecting the execution to degrade the performance”, or “Why is allocating more servers not improving the execution time?”. The objective of the proposed research is to develop automated solutions that address this increasingly difficult challenge of performance troubleshooting and tuning of cloud based DDDAS applications by synergistically exploiting techniques from hierarchical performance modeling, on-demand execution tracing, and discriminative sequence mining. By leveraging theoretical performance modeling for scalable diagnosis, the proposed research will develop new scalable techniques for troubleshooting large-scale systems, reducing the troubleshooting time and learning curve for system administrators, which allow IT personnel to focus and spend more time on other value-added tasks.