The goal of this project is to research a full-system, cycle-accurate computer simulator that runs approximately three orders of magnitude faster than the fastest current simulators of similar accuracy. Such performance is achieved by a careful partitioning of the simulator that permits the most computationally intensive component to be efficiently implemented in reconfigurable hardware. This partitioning also simplifies and abstracts the simulators, making them easier to make more complete and accurate. For example, an early prototype boots unmodified Linux. The goals of this project are to (i) build and evaluate instances of such simulators, (ii) refine tools to build and use such simulators to the point that they could be used in an advanced undergraduate class and (iii) to further optimize the performance of such simulators by improving the partition interface and implementation strategies. Since the resulting simulators are simultaneously accurate, fast and full-system, we anticipate that all users of cycle-accurate computer simulators, ranging from students to architects designing computers to engineers developing the computers to advanced end-users who wish to tune their applications for a particular computer, will eventually use a form of such a simulator. Sharing the same infrastructure will likely lead to better communication between these groups, resulting in better computer systems overall. We expect to disseminate versions of the simulators and tools primarily through a web interface that will allow users to specify a simulator and run it over the web at our site.
