Loading...

Processor Allocation for Future Multi-Core Chip-Multiprocessor

Agha Ali Akbari, Fatemeh | 2014

782 Viewed
  1. Type of Document: M.Sc. Thesis
  2. Language: Farsi
  3. Document No: 46429 (19)
  4. University: Sharif University of Technology
  5. Department: Computer Engineering
  6. Advisor(s): Sarbazi Azad, Hamid
  7. Abstract:
  8. Since 2005, processor designers have increased core counts to exploit Moore’s Law scaling, rather than focusing on single-core performance. For decades, this approach provides desired performance for parallel and multithreaded workloads. On the other hand, rising of utilization wall limits the number of transistors that can be powered on in chip and result in a large region to be dark. So, same as before trend for performance scaling in future multi processor, an appropriate architecture is essential. There are some structures for this era which used specialization approach to cope with the limited power budget. Therefore, in this thesis, we propose a general-purpose platform that provides performance demands with fine-grained power management ability. With regard to vital role of network-on-chip in improving performance and power consumption, we first reveal challenges and opportunities of designing this platform in many-core chip-multiprocessors. We then discuss this proposed idea for interconnection networks. Furthermore, by using clustered mesh and sharing resource capability, we can improve performance for applications with different needs. Furthermore, our platform considers multi-threaded levels for processing cores in order to provide cores with different computational capability. It can also manage the power consumption with arbitrary granularity. Besides importance of this architecture, efficient processor allocation is crucial for obtaining high performance in parallel and distributed systems. Processor allocation algorithms choose set of processors for incoming jobs. According to extreme resources and limited power budget in dark silicon age, new dimensions are added to processor allocation. So, an adequate algorithm must reduce run time without increasing power while providing suitable thermal distribution to avoid hot spots. In this line, we can improve performance through using extra resources, if power budget permitted us. Considering this idea in processor allocation provides multi-programmed speedup. Therefore, proposed algorithm with these features named as Dark Silicon Aware Processor Allocation (DSAPA). Compared to User-Aware and Topological Sprinting approaches, our results reveal that 36% and 30% improvement in speedup and by up to 33% and 36% improvements in power consumption, respectively
  9. Keywords:
  10. Efficiency ; Power Consumption ; Network-on-Chip (NOC) ; Multiprocessor System ; Multicore Processors ; Processor Allocation ; Future Chip Multiprocessors ; Dark Silicon ; General purpose Platform

 Digital Object List

 Bookmark

No TOC