Embedded systems are migrating to multicore chips. Where a system’s software can be contained on a single core, impact may be minimal. Where it must be spread over multiple cores, the impact could be significant. This migration is motivated by a cost barrier that prevents further CPU acceleration. Increasing the clock frequency requires an increase in power, and therefore, heat. Currently, any further increase in heat would require prohibitively expensive cooling mechanisms. The only financially viable means of improving embedded processor performance is to add CPUs to the chip. This allows chip makers to reduce clock speed, allowing reduction of size, weight, power, and cost. It also reduces the likelihood that a single core would have the throughput to support a system’s software. Pieces of code that otherwise would have run sequentially may run in parallel. This presentation will describe concurrency problems this may create, and offer possible mitigating actions.
|
