简介：Asatypicalerasurecodingchoice,Reed-Solomon（RS）codeshavesuchhighrepaircostthatthereisapenaltyforhighreliabilityandstorageefficiency,therebytheyarenotsuitableingeo-distributedstoragesystems.Wepresentanovelfamilyofconcurrentregenerationcodeswithlocalreconstruction（CRL）inthispaper.TheCRLcodesenjoythreebenefits.Firstly,theyareabletominimizethenetworkbandwidthfornoderepair.Secondly,theycanreducethenumberofaccessednodesbycalculatingparitiesfromasubsetofdatachunksandusinganimpliedparitychunk.Thirdly,theyarefasterthanexistingerasurecodesforreconstructioningeo-distributedstoragesystems.Inaddition,wedemonstratehowtheCRLcodesovercomethelimitationsoftheReed-Solomoncodes.Wealsoillustrateanalyticallythattheyareexcellentinthetrade-offbetweenchunklocalityandminimumdistance.Furthermore,wepresenttheoreticalanalysisincludinglatencyanalysisandreliabilityanalysisfortheCRLcodes.Byusingquantitycomparisons,weprovethatCRL（6,2,2）isonly0.657xofAzureLRC（6,2,2）,wheretherearesixdatachunks,twoglobalparities,andtwolocalparities,andCRL（10,4,2）isonly0.656xofHDFS-Xorbas（10,4,2）,wherethereare10datachunks,fourlocalparities,andtwoglobalparitiesrespectively,intermsofdatareconstructiontimes.OurexperimentalresultsshowtheperformanceofCRLbyconductingperformanceevaluationsinbothtwokindsofenvironments：1）itisatleast57.25%and66.85%morethanitscompetitorsintermsofencodinganddecodingthroughputsinmemory,and2）ithasatleast1.46xand1.21xhigherencodinganddecodingthroughputsthanitscompetitorsinJBOD（JustaBunchOfDisks）.WealsoillustratethatCRLis28.79%and30.19%morethanLRConencodinganddecodingthroughputsinageo-distributedenvironment.