Wireless Platform

The rapid growth in wireless technology has exposed a number of problems. Rapid growth in available bandwidth to the extent that wireless devices can communicate at speeds a few generations behind traditionally wired protocols (ethernet etc.) has resulted in early standards being rapidly overtaken. Consequently, there is a need to interface to multiple standards transparently yet, at the same time, many wireless designs are highly memory-limited as well as being highly sensitive to power consumption.

This combination of problems results in a need to do a systematic investigation of wireless needs, with a goal of establishing a common platform.

Thoughts and ideas on these pages originate from discussions in the Advanced Computer Architecture Laboratory at the University of Michigan. This site represents a range of projects from different groups. Watch this space for new participation.

Potential Projects

This partial list of projects represents a mix of work in progress and potential future projects. Further clarification will be posted as projects progress.

• mainstreaming DSP
  Should digital signal processing be done on special-purpose processors, or integrated with a standard CPU? Some issues:
  • multiply-accumulate (MAC), at the heart of many DSP kernels, would be simple to add to a conventional instruction set – would the performance gain be worth it? How would the performance compare with a DSP-specific design?
  • what are suitable benchmarks? There is no equivalent of SPEC in the DSP world, so a standard benchmark suite needs to be developed
• low-energy computing
  How can power requirements of computers be reduced? Can aggressive performance demands be met in more energy-efficient ways with modest compromise on performance. Some exist sing projects:
  • look for “low power” at <http://www.eecs.umich.edu/~tnm/papers.html> – detail to be added
• operating systems issues
  given the small memory footprint and realtime requirements, what is a suitable operating system? What modifications may be needed to suit this environment? Some issues:
  • proprietary vs. free – the free software route has a lot of attractions, not least avoiding the possibility of a dominant player undermining standards; Linux already has some penetration (e.g. cell phones in China and Japan) but other options are possible (e.g., ); the major proprietary competition is Symbian
  • fine-grained protection – see virtual protocol stacks for motivation; we will likely not be able to justify the overheads of VM, but protection of some sort is desirable
• network protocols
  The proliferation of protocols representing different bandwidth requirements, different area coverage and different price points makes interoperability difficult – should a device support Bluetooth if it supports IEEE 802.11g? Should it be able to discover new protocols on the fly? Some issues:
  • soft radio – to what extent can traditionally analog functionality be moved into software to increase flexibility? What can advances in computer architecture contribute to making soft radio more practical?
  • virtual protocol stacks – what are issues in finding a new protocol over a wireless link? Can we “page in” a new protocol stack? How can we be sure the code is trustworthy? Can we implement protection on a granularity to suit isolating the protocol stack from critical data?

While many of these project ideas go outside architecture, they have strong links to architecture. Efficient implementation may depend on issues like memory hierarchy and CPU design. The initial focus therefore is on architecture issues. However, expansions to networks and operating systems is possible.

People

Why a Koala?

A koala, often mistakenly called a “bear”, is a highly energy-efficient marsupial. Shown here in repose in its pose for hot weather, it flops out over a branch catching a cooling breeze. In cold weather, it coils up into a fork in a tree to conserve body heat. A koala only eats leaves of a very specific variety of eucalyptus, which diet does not yield much energy. Like a mobile computing device, a koala has to have energy-efficient habits to be competitive.

Links