As part of CCC’s ongoing support of visioning workshops, Josep Torrellas (University of Illinois at Urbana-Champaign) and Mark Oskin (University of Washington) co-organized the second workshop in a series on Advancing Computer Architecture Research. This workshop — Laying a New Foundation for Information Technology: Computer Architecture for 2025 and Beyond — was held on the campus of the University of Washington on September 20-21, 2010.
Computer Architecture Research as a Foundation for IT
The degree to which information technology has impacted society cannot be understated, and computer architecture has played a foundational role in enabling that revolution. Computer architecture has been the engine for translating Moore’s Law technology improvements into improved software performance for nearly 50 years. Increasing transistor budgets have enabled increasingly sophisticated processor implementations, and by virtue of an abstraction layer between the software and the hardware — the instruction set architecture — the resulting performance benefits have been available with little if any effort from the programmer or end user.
Historically, the computing industry has been driven by a set of exponential increases in single-thread performance. Increased clock rates have delivered much of this performance, but they have now stalled. Multi-core processors have appeared, promising a continuation of exponential performance increases, but this forces the mainstream adoption of parallel processing. These are truly seismic forcing events for computer architecture.
Envisioning the Future of the Field
With computer architecture being so central to information technology’s success, it is important for the computer architecture research community to envision future needs and trends and respond to them. Computer architecture currently faces a turning point that calls for dramatically new techniques and approaches. This turning point is caused by several key trends:
– Technology Drivers: On one hand, technology trends will change the shape of future computer architectures. In particular, the long-discussed slowing in Moore’s Law improvements and Denard scaling makes it increasingly difficult to extract more computer performance by employing more transistors on-chip. Power limits and reduced semiconductor reliability will make future scaling more difficult to leverage.
– Application Drivers: Computer architectures must also respond to the changing nature of applications and software. Recent years have seen the emergence of rich multimedia applications and data-intensive computing, as well as extensive growth in both mobile/embedded computing at one extreme, and warehouse-scale data center computing at the other. All these call for rethinking the simple architectural interfaces (load, store, compute) that have remained largely unchanged for decades.
– Metrics and Goals: A final important trend calling for architectural attention concerns the metrics and goals by which computer systems are evaluated. While computing has until recently been almost exclusively performance-focused, it is now abundantly clear that other metrics demand attention. In addition to performance, computer systems are expected to be reliable, to be secure, and to operate with manageable and malleable power budgets. Achieving these goals requires extensive hardware and software cooperation, and thus it becomes vital that the computer architecture be designed with appropriate mechanisms and interfaces to facilitate this cooperation.
The Visioning Workshop
Torrellas and Oskin brought together 20 researchers from the U.S. and E.U. to discuss these problems in order to help formulate suggested research topics for the architecture community. A report as well as a set of presentations will be posted on the CCC website shortly.
In the meantime, please check out the report from the first ACAR visioning workshop, held earlier this year: http://cra.org/ccc/docs/ACAR_Report_Popular-Parallel-Programming.pdf. Slides from that meeting will soon be available here.
(Contributed by Bill Feiereisen, Lockheed Martin Corp., on behalf of the workshop participants)
