Simple and Efficient Instrumentation for the DSZOOM System

Abstract

An efficient and robust instrumentation tool (or compiler support) is necessary for an efficient implementation of fine-grain software-based shared memory systems (SW-DSMs). The DSZOOM system, developed by the Uppsala Architecture Research Team (UART) at Uppsala University, is a sequentially consistent SW-DSM originally developed using EEL (Executable Editing Library) - a binary modification tool from University of Wisconsin-Madison. In this thesis, we identify several weaknesses of this original approach and present a new and simple tool for assembler instrumentation. This tool can instrument (modify) highly optimized compiler output for the newest UltraSPARC processors. Currently, the focus of the tool is load-, store-, and load-store instrumentation. We also present several low-level optimization techniques that significantly improve the performance of the DSZOOM system. One of the presented techniques is a store-buffer register optimization, a latency-hiding mechanism for memory-store operations, that can lower instrumentation overheads for some applications (as much as 45% for LU-cont, running on two nodes with 8 processors each). We also show that by using this new DSZOOM system we execute faster than the old one on all applications in the SPLASH-2 benchmark suite. Improvements range from 1.07 to 2.82 times (average 1.73).

Available as PDF (1.0 MB)

BibTeX file entry: Grenholm:2002:dec