1 Introduction and main contributions As so many machines are clustered in datacenters,failures occur frequently[1,2].Redundancy schemes,like replication and erasure coding,are deployed to store redundant data to tole...1 Introduction and main contributions As so many machines are clustered in datacenters,failures occur frequently[1,2].Redundancy schemes,like replication and erasure coding,are deployed to store redundant data to tolerate failures.For distributed systems with these two redundancy schemes,new data are hot in replication,and usually turn cold as time goes by.An encoder performing the conversion can cause large overheads on network and put data in low reliability when the encoder is not well designed.展开更多
基金supported in part by NSFC(61832020)National Key R&D Program of China(2018YFB10033005)+2 种基金Hubei Province Technical Innovation Special Project(2017AAA129)Wuhan Application Basic Research Project(2017010201010103)Fundamental Research Funds for the Central Universities.
文摘1 Introduction and main contributions As so many machines are clustered in datacenters,failures occur frequently[1,2].Redundancy schemes,like replication and erasure coding,are deployed to store redundant data to tolerate failures.For distributed systems with these two redundancy schemes,new data are hot in replication,and usually turn cold as time goes by.An encoder performing the conversion can cause large overheads on network and put data in low reliability when the encoder is not well designed.
