Due to 5G's stringent and uncertainty traffic requirements,open ecosystem would be one inevitable way to develop 5G.On the other hand,GPP based mobile communication becomes appealing recently attributed to its str...Due to 5G's stringent and uncertainty traffic requirements,open ecosystem would be one inevitable way to develop 5G.On the other hand,GPP based mobile communication becomes appealing recently attributed to its striking advantage in flexibility and re-configurability.In this paper,both the advantages and challenges of GPP platform are detailed analyzed.Furthermore,both GPP based software and hardware architectures for open 5G are presented and the performances of real-time signal processing and power consumption are also evaluated.The evaluation results indicate that turbo and power consumption may be another challengeable problem should be further solved to meet the requirements of realistic deployments.展开更多
The many-accelerator architecture, mostly composed of general-purpose cores and accelerator-like function units （FUs）, becomes a great alternative to homogeneous chip multiprocessors （CMPs） for its superior power-...The many-accelerator architecture, mostly composed of general-purpose cores and accelerator-like function units （FUs）, becomes a great alternative to homogeneous chip multiprocessors （CMPs） for its superior power-efficiency. However, the emerging many-accelerator processor shows a much more complicated memory accessing pattern than general purpose processors （GPPs） because the abundant on-chip FUs tend to generate highly-concurrent memory streams with distinct locality and bandwidth demand. The disordered memory streams issued by diverse accelerators exhibit a mutual- interference behavior and cannot be efficiently handled by the orthodox main memory interface that provides an inflexible data fetching mode. Unlike the traditional DRAM memory, our proposed Aggregation Memory System （AMS） can function adaptively to the characterized memory streams from different FUs, because it provides the FUs with different data fetching sizes and protects their locality in memory access by intelligently interleaving their data to memory devices through sub-rank binding. Moreover, AMS can batch the requests without sub-rank conflict into a read burst with our optimized memory scheduling policy. Experimental results from trace-based simulation show both conspicuous performance boost and energy saving brought by AMS.展开更多
基金funded in part by National Natural Science Foundation of China(grant NO.61471347)National S&T Mayor Project of the Ministry of S&T of China(grant NO.2016ZX03001020-003)+1 种基金key program for international S&T Cooperation Program of China(grant NO.2014DFA11640)Shanghai Natural Science Foundation(grant NO.16ZR1435100)
文摘Due to 5G's stringent and uncertainty traffic requirements,open ecosystem would be one inevitable way to develop 5G.On the other hand,GPP based mobile communication becomes appealing recently attributed to its striking advantage in flexibility and re-configurability.In this paper,both the advantages and challenges of GPP platform are detailed analyzed.Furthermore,both GPP based software and hardware architectures for open 5G are presented and the performances of real-time signal processing and power consumption are also evaluated.The evaluation results indicate that turbo and power consumption may be another challengeable problem should be further solved to meet the requirements of realistic deployments.
基金Supported by the National Natural Science Foundation of China under Grant Nos.61173006,60921002the National BasicResearch 973 Program of China under Grant No.2011CB302503the Strategic Priority Research Program of the Chinese Academyof Sciences under Grant No.XDA06010403
文摘The many-accelerator architecture, mostly composed of general-purpose cores and accelerator-like function units （FUs）, becomes a great alternative to homogeneous chip multiprocessors （CMPs） for its superior power-efficiency. However, the emerging many-accelerator processor shows a much more complicated memory accessing pattern than general purpose processors （GPPs） because the abundant on-chip FUs tend to generate highly-concurrent memory streams with distinct locality and bandwidth demand. The disordered memory streams issued by diverse accelerators exhibit a mutual- interference behavior and cannot be efficiently handled by the orthodox main memory interface that provides an inflexible data fetching mode. Unlike the traditional DRAM memory, our proposed Aggregation Memory System （AMS） can function adaptively to the characterized memory streams from different FUs, because it provides the FUs with different data fetching sizes and protects their locality in memory access by intelligently interleaving their data to memory devices through sub-rank binding. Moreover, AMS can batch the requests without sub-rank conflict into a read burst with our optimized memory scheduling policy. Experimental results from trace-based simulation show both conspicuous performance boost and energy saving brought by AMS.
