PIM-Based Introspection

Hans P. Zima, JPL, California Institute of Technology, and Institute for Software Science, University of Vienna, Austria

Abstract

Processing-In-Memory (PIM) architectures avoid the von Neumann bottleneck by integrating high-density DRAM and CMOS logic on the same chip. Parallel systems based on this new technology are expected to provide higher scalability, adaptability, robustness, fault tolerance and lower power consumption than current MPPs or commodity clusters. In this talk we discuss the capability of PIM-based architectures to create a large number of mobile lightweight threads that can be efficiently managed via special hardware support. We focus on the use of such threads for the implementation of asynchronous agents that perform introspective monitoring of application programs. Introspection can be put to many uses, including dynamic program validation, analysis of program execution behavior and intrusion detection and prevention. In a case study, we will discuss the use of introspection for performance monitoring, analysis, and feedback-oriented performance tuning.