Application of full-system simulation in exploratory system design and development

IBM Journal of Research and Development, Mar-May 2006 by Peterson, J L, Bohrer, P J, Chen, L, Elnozahy, E N, Et al

This paper describes the design and application of a full-system simulation environment that has been widely used in the exploration of the IBM PowerPC� processor and system design. The IBM full-system simulator has been developed to meet the needs of hardware and software designers for fast, accurate, execution-driven simulation of complete systems, incorporating parameterized architectural models. This environment enables the development and tuning of production-level operating systems, compilers, and critical software support well in advance of hardware availability, which can significantly shorten the critical path of system development. The ability to develop early versions of software can benefit hardware development by identifying design issues that may affect functionality and performance far earlier in the development cycle, when they are much less costly to correct. In this paper, we describe features of the simulation environment and present examples of its application in the context of the Sony-Toshiba-IBM Cell Broadband Engine(TM) and IBM PERCS development projects.

Introduction

Recent trends in processor technology suggest that future gains in system performance will rely increasingly on integrated solutions that leverage improvements across many system components, such as memory architectures, interconnect technologies, accelerators, operating systems, compilers, and application libraries. This development has led to increased design complexity in individual system components, in the introduction of new components, and in new component interactions. This growth in complexity has placed a new emphasis on the simulation of system behavior and characteristics as a key aspect of system exploration and design. To make effective design decisions, software developers and system architects require tools that assist in quantifying the merits and limitations of specific hardware and software architecture alternatives [1]. As interactions among processors and system components become increasingly important in system design, full-system simulation has become an indispensable tool for the evaluation of new systems.

The IBM full-system simulator, internally referred to as "Mambo" [2], has been developed and refined in conjunction with several large system design projects built on the IBM Power Architecture* [3]. The POWER* technology is an instruction set architecture (ISA) that spans applications from consumer electronics to supercomputers. Power Architecture encompasses PowerPC*, POWER4*. and POWERS* processors. As one of many full-system simulators [4-6], Mambo allows a full operating system (OS), such as Linux**, to run interactively in the simulation, thereby furnishing a complete environment for applications that require interprocess or complex OS interactions. In addition, a standalone environment is provided for self-contained applications. In this environment, Mambo intercepts and marshals the application system calls to the underlying host to optimize execution. Mambo itself runs atop a variety of host OS and platform combinations. Figure 1 shows the Mambo execution and simulation stack.

Mambo facilitates experimentation with and evaluation of a wide variety of system elements, including ISAs, address translation mechanisms, memory coherence protocols, hypervisor software and hardware support, hardware acceleration engines, and microarchitectural features and resource allocation. Mambo performance models may be driven by instruction or memory traces. Environment flexibility enables users to configure a simulation platform based on customer-driven application and analysis requirements.

Mambo is particularly useful in evaluating new designs that involve complex interactions among several different system components. Mambo can support detailed evaluation and testing of complex hardware interactions, such as locking and cache coherence protocols. Additionally, Mambo enables the development and tuning of production-level operating systems, compilers, and substantial application software before hardware prototypes are available. By facilitating the concurrent development of hardware and system software, Mambo can significantly shorten the critical path of development schedules. This early software development activity can also benefit the hardware development effort by identifying design issues that have an impact on functionality or performance far earlier in the development cycle, when the cost to fix such issues is considerably less.

Technology overview and design techniques

A Mambo simulation accurately models an entire system of processing elements, devices, interconnect topology, and protocols. By virtue of its execution-driven capabilities, it simulates core instruction execution and the subtleties of interactions with surrounding system components. At the heart of a Mambo simulation environment is its processor model. Mambo provides complete, bit-accurate models of several different PowerPC processors [7], ranging from the embedded 405GP and 750 processors to the PowerPC 970 (used in the Apple G5 systems and PowerPC-based blades) to the new Sony-Toshiba-IBM (STI) Cell Broadband Engine** (Cell BE) processor (used for the Sony PlayStation** 3). The Power Architecture defines the basic instructions and registers and their operation. The Mambo processor model simulates the entire instruction set, including both user and privileged-state operations, and optional features defined by the architecture, such as 32-bit and 64-bit modes and floating-point extensions. Each simulated processor is defined as a collection of these features.