Unvalidated Simulations and the Quantitative Approach to Computer Architecture

Friday, November 5th 2010, 10:30am
302 Building, Room 308


Most papers today include quantitative results which are based on simulations which ignore some details of the architecture that the designer/experimenter regards as unnecessary. It is rare to see a validation of the simulation results against real machines or cycle-accurate simulations of the proposed machine. We present a case study to show how performance estimates of a single-core, in-order PowerPC pipeline get affected when the pipeline details are abstracted away in simulation. In the context of evaluating three different branch predictors, we compare the accuracy of two abstract simulators against cycle-accurate simulators which capture all the pipeline details. For each design with a different branch predictor we build a cycle-accurate simulator to run on an FPGA platform. Our studies show that sometimes abstract simulators produce results which are wrong even qualitatively.

This talk is based on the IBM-MIT joint project to model PowerPC involving Asif Khan and M. Vijayaraghavan of MIT and K. Ekanadham and Jessica Tseng of IBM Research.

Speaker Bio

Arvind is the Johnson Professor of Computer Science and Engineering at MIT. Prof. Arvind also serves as the SNU-WCU Distinguished Professor of Computer Science and Engineering at SNU. Arvind's group, in collaboration with Motorola, built the Monsoon dataflow machines and its associated software in the late eighties. In 2000, Arvind started Sandburst which was sold to Broadcom in 2006. In 2003, Arvind co-founded Bluespec Inc., an EDA company to produce a set of tools for high-level synthesis. In 2001, Dr. R. S. Nikhil and Arvind published the book "Implicit parallel programming in pH". Arvind's current research focus is on enabling rapid development of embedded systems.

Arvind is a Fellow of IEEE and ACM, and a member of the National Academy of Engineering.