• View
• Edit
• History
• Print

Overview

The Jamaica project is investigating the design of chip multi-processors (CMPs) and their accompanying parallel software environments. CMP architectures have been widely accepted by many processor chip manufacturers as a solution to the design problems accompanying the scale-up to "billion transistor" chips. The rationale for this choice is that designing logic to interconnect multiple cores based on existing designs is enormously simpler than trying to build one core which will use all the available silicon.

Despite the widespread convergence on CMP as a promising design strategy, there are many issues yet to be resolved in the design of both the hardware and its accompanying software environment. In particular:

• The hardware must be able to efficiently support an operating system which can distribute

execution of application code to all the available cores. Distribution and synchronization costs must be low and it must be easy to detect the presence of idle cores.

• The operating system must, in turn, rely upon advanced compiler technology to automate,

as far as possible, this distribution of work. Most applications cannot feasibly be designed (or rewritten) to allow for all possible CMP configurations, coping with anything from, say, 2 to 64 cores.

• A dynamic parallelizing compiler is essential if the distribution problem is to be

addressed. Both application and operating system code need to be optimized appropriate to the CMP configuration found at runtime or even recompiled on the fly using feedback directed recompilation.

All three of these elements of the CMP platform need to be designed together if the promise of CMP architectures is to be fulfilled.

The Jamaica project is organized into three collaborating strands reflecting the interplay between computer architecture, compiler and operating system. (Quick links: Project members, Publications, Intranet (campus only))