There is a signup schedule for this event (UT EID requir

ed).

Type of Talk:  UTCS & ECE FACULTY CANDIDATE

Speaker/Affiliation:  Engin Ipek/Microsoft Research

Date/Time:  Wednesday, April 15, 2009 2:00 p.m.

Locatio

n:  ACES 6.304

Hosts: Steve Keckler and Yale Patt

Talk Title:
"Reconfigurable and Self-Optimizing Multicore Arc

hitectures"

Talk Abstract:
As industry rides the tran

sistor density growth in multicore processors by providing more and more co

res, these will exert
increasing levels of pressure on shared system
resources. Efficient resource management becomes critical to obtaining
high utilization, and eliminating potential bandwidth, latency, and c

ost barriers in multicore systems. Unfortunately, current
hardware p

olicies for microarchitectural resource management are ad hoc at best, and
are generally incapable of providing basic
functionalities like anti

cipating the long-term consequences of scheduling decisions (planning), or
generalizing from experience
obtained through past resource allocati

on decisions to act successfully in new situations (learning). As a result

, current hardware controllers tend to grossly underutilize the (already li

mited) platform resources available. In this talk, using the problem of me

mory scheduling as context, I will describe the use of machine learning (M

L) technology in designing self-optimizing, adaptive hardware controllers

capable of planning, learning, and continuously adapting to changing work

load demands. An ML-based design approach allows the hardware designer to f

ocus on what performance target the controller should accomplish and what s

ystem variables might be useful to ultimately derive a good control policy

, rather than devising a fixed policy that describes exactly how the contro

ller should accomplish that target. This not only eliminates much of the hu

man design effort involved in traditional controller design, but also yiel

ds higher performing, more efficient controllers.

This work was
completed as part of Engin Ipek''s Ph.D. thesis at Cornell''s Computer Sys

tems Laboratory. It has been nominated for the 2008 ACM Doctoral Dissertati

on Award by Cornell University.