Research And Implementation Of Firmware Architecture In Massive Parallel Computer

Firmware is one of the key components of Massive Parallel Processing systems based on IA64. As far as the function is concerned, firmware is more powerful than the traditional BIOS. Firmware performs booting initialization configuration and self-testing of a single processing node. Besides, it also performs the manipulation of multiple-processor, booting operating system, and driving IO devices. While the system is running, firmware takes the responsibility for run-time called procedure, error detecting and recovery, and system diagnosis, etc. In a word, firmware plays a key role in a MPP system. During the time from the system is powered on to OS taking over the control of the system, the whole system is under the control of firmware, which provides a pre-boot environment. Firmware is able to access the most resource of the whole system. For convenience, firmware also provides users with a kind of interface and some commands. So, firmware can be regarded as a simple operating system.Firmware does not retire after the operating system has taken over the control of the whole system. During the run time, firmware provides the operating system with some run-time services and system-calls. At the same time, firmware also takes corresponding actions for events of the platform, including all kinds of exceptions and traps. In order to improve the system's reliability, firmware takes the responsibility for error detecting and recovery. For those recoverable errors, it is transparent to the operating system.According to the requirement of MPP system based on CCNUMA, this paper presents a full firmware architecture, which performs the booting of each processing node. At the same time, this firmware provides a operating system visible abstraction layer of the lower level hardware platform. Besides, this firmware implements a well booting flow with high efficiency. In order to enhance the system's performance, this firmware improves the runtime event processing and run-time service. For scalability, this firmware presents a method for processing node's unloading and plugging in.Firmware is of key importance in the MPP systems. The firmware architecture presented in this paper is feasible and workable. So, the result of this paper is very significant for the design and implementation of MPP systems.