Simultaneous multithreading: Operating system perspective

Developing CPU architecture is a very complicated, iterative process that requires significant time and money investments. The motivation for this work is to find ways to decreases the amount of time and money needed for the development of hardware architectures. The main problem is that it is very difficult to determine the performance of the architecture, since it is impossible to take any performance measurements untill upon completion of the development process. Consecutively, it is impossible to improve the performance of the product or to predict the influence of different parts of the architecture on the architecture\u27s overall performance. Another problem is that this type of development does not allow for the developed system to be reconfigured or altered without complete re-development. . The solution to the problems mentioned above is the software simulators that allow researching the architecture before even starting to cut the silicon.. Simultaneous multithreading (SMT) is a modern approach to CPU design. This technique increases the system throughput by decreasing both total instruction delay and stall times of the CPU. The gain in performance of a typical SMT processor is achieved by allowing the instructions from several threads to be fetched by an operating system into the CPU simultaneously. In order to function successfully the CPU needs software support. In modern computer systems the influence of an operating system on overall system performance can no longer be ignored. It is important to understand that the union of the CPU and the supporting operating system and their interdependency determines the overall performance of any computer system. In the system that has been implemented on hardware level such analysis is impossible, since the hardware system is neither flexible nor configurable. However, in the SMT architecture, the system is capable of performing some useful work even if a task has generated an error. A wide range of simulators is described in the literature, and a lot of them are publicly accessible. The main goal of this work is to modify an existing SEVIOS/Topsy simulator to achieve a simple, configurable, publicly accessible SMT SEVIOS/Topsy simulator that must also include an SMT Topsy.. The simulator should demonstrate the fetching process of the SMT MIPS, as well as scheduling aspects of the CPU and the operating system integrated environment.. This work covers a broad range of aspects, among which are: 1) Completion of SMT MIPS and SMT Topsy specifications; 2) Integration of MXS into SIMOS/Topsy; 3) Modifications to the fetching unit of MXS that allow to support SMT; 4) Addition of SMT support to Topsy;; This work uses Topsy/R4000 simulator developed at Swiss Federal Institute of Technology, and the MXS (R10000) part of the SimOS simulator developed at Stanford University. Development process utilizes C high-level language, Intel and MIPS assembly languages. The result of this work is a development of a complete computer system software simulator. The simulator allows taking performance measurements and reconfiguration of SMT Topsy and the fetching unit of the SMT MXS. The simulator is modular: that is any of its parts can be substituted with other parts that perform similar functionality. It also means that the whole simulator can be integrated into a larger scale simulation project. The development of this simulator significantly decreases the amount of time and money needed for the development of hardware architectures and provides new ways in researching the influence of an operating system on the performance of the computer system as a whole