NGS: Performance Mining of Large-Scale Data-Intensive Distributed Object Applications

Rensselaer Polytechnic Institute, Troy NY

Investigators

Abstract

EIA-0103708 Mohammed J. Zaki Rensselaer Polytechnic Institute NGS Performance Mining of large-scale Data-Intensive Distributed Object Applications The objective of the proposal is to develop a performance measurements-based run-time environment for supporting large data-intensive distributed object applications. The system will provide continuous and adaptive performance optimization via a combination of performance data mining, critical path discovery and speculative execution. To address these challenges for next generation software systems we propose to develop the PERFMINER engine for the performance mining. PERFMINER, a system for continuous performance optimization via mining, will enable a distributed object system to: 1) discover its own critical path, 2) detect new opportunities for speculative processing, and 3) to facilitate modifying an object's behaviors (i.e., methods) at run-time in response to newly acquired knowledge.