We propose a new approach of performing automatic program parallelization: 
divide a program into traces and execute these traces in parallel.  We 
examine the issues that need to be addressed for this approach to be 
successful. These issues include how traces are collected, how they are 
selected, how they are scheduled, and how dependence is enforced.  We also 
present a feasibility study that shows that our approach has merit.