Shenzhen Advanced Institute has made progress in the study of the characteristics of blockchain benchmarking procedures

[ Instrument Network Instrument Development ] Recently, the Institute of Advanced Technology of Shenzhen Institute of Advanced Science and Technology of the Chinese Academy of Sciences has made progress in the feature mapping and performance optimization research of blockchain benchmarking procedures. The related results are BBS: Micro-architecture Benchmarking Blockchain Systems The through Machine Learning and Fuzzy Set is received by the IEEE International Symposium on High-Performance Computer Architecture (HPCA) 2020. The conference is dedicated to presenting and discussing new ideas and latest research in computer architecture. Zhu Liang, a master student of the Shenzhen Institute of Advanced Intelligence Computing Center, is the first author of the paper. Chen Chao is the second author of the thesis, and researcher Yu Zhibin is the author of the communication.

The research aims to characterize the blockchain system at the processor microarchitecture layer, analyze the reasons for the poor performance of the blockchain from the microarchitecture level and propose an optimization scheme. Since the current blockchain benchmarking program only focuses on the performance indicators of the blockchain as a whole (such as throughput, delay, etc.), it does not characterize the blockchain system from the microarchitecture layer, which makes it difficult for people to locate the zone. The reason for the poor performance of the blockchain system. Therefore, people don't know which existing CPU micro-architecture should be chosen, or how to design the CPU's micro-architecture to run the blockchain system more efficiently. The BBS compares the microarchitecture events in the running process by collecting the blockchain system benchmarks, and sorts the importance of these events to compare the characteristics of different blockchain systems. By studying the results of the analysis, it is possible to clearly show the reasons for the poor performance of the blockchain system. At the same time, the BBS selects the results of the order importance of the blockchain microarchitecture events by using fuzzy mathematics, and selects an appropriate number of events important to the performance of the benchmark program. Then, when characterizing the microarchitecture layer of the blockchain, only a small number of events need to be observed to comprehensively and accurately measure the performance of the blockchain system. The selected events in the research process can also analyze the similarity of the existing blockchain benchmarks, and filter out the blockchain benchmarks with too high similarity to reduce the test overhead. Experiments show that in eight high-performance server cluster environments, BBS can not only increase the throughput of Hyperledger Fabric by 70%, but also reduce its latency by 55%.

The above work was funded by the key research and development program “Software-Defined Cloud Computing Resource Management” and the National Natural Science Foundation.