Community Research and Development Information Service - CORDIS


MECCA Report Summary

Project ID: 340328
Funded under: FP7-IDEAS-ERC
Country: Sweden

Mid-Term Report Summary - MECCA (Meeting Challenges in Computer Architecture)

The overarching goal of MECCA is to make significant advances in how future computers use available computational resources to meet challenging goals concerning computational performance, energy consumption and (time) predictability. MECCA is confronted with three major challenges anchored in barriers related to the shift to parallel architectures in the beginning of the millenium. To address the three challenges concerning performance, power and predictability, MECCAs unique approach has been to envision a resource management approach that is guided by both static semantic information from the programming model level as well as dynamic information from the architecture level to meet the goals. In essence, the focus has been to show that it is possible to solve pressing problems in a new way by allowing layers in the computational stack to exchange information.

Concerning parallelism management, the project has developed several unconventional and novel approaches to use computational resources more efficiently. Specifically, the project has focused on on-chip cache hierarchies as most of the on-chip resources are used for caching and has a significant impact on performance and power consumption. To this end, a new compression method has been developed that can utilize cache capacity a factor of three more efficiently. The technology has been transferred to a startup company for commercialization. The project has also developed a novel new concept for cache management in which the runtime system effectively identifies and evicts so called dead blocks; content of the cache that is not used. This concept uses information gathered from both the programming model as well as the architecture to allow for substantially improved management as compared to existing techniques.

Regarding power management, this project has pursued an unconventional approach in which it is assumed that computations are associated with computational deadlines. A progress-tracking method has been developed that can accurately keep track of how much time slack there is. This time slack can be used to use slimmed resources for the computational task by which power can be saved. The results show that substantial power savings of at least a factor of two can be achieved by the progress tracking and scheduling methods developed depending on the amount of time slack.

Finally, as for predictability management, an open problem is how to enjoy speedup of parallel algorithms on modern parallel architectures when stringent timing requirements must be met such as in safety-critical applications, eg vehicles. The issue is that when guaranteeing real-time requirements, system resources are easily over-provisioned making it favorable to run applications sequentially rather than in parallel. For the first time, this project has developed a scheduling approach that allows tasks in a parallel algorithm to be scheduled dynamically and at the same time avoid timing anomalies that would make them miss computational deadlines. A developed timing-anomaly-free dynamic scheduling method is shown to reduce the guaranteed makespan, that is, the worst-case execution time of the parallel algorithm substantially.

Reported by

Follow us on: RSS Facebook Twitter YouTube Managed by the EU Publications Office Top