A Quantitative Analysis oftheUtilization ofAssociative Memories inDataManagement

An associative approach todatamanagementisde- veloped, anda methodology formeasuringtheperformance and effectiveness ofthisapproach isprovided. Mathematical models whichquantitatively describe retrieval, storage requirements, up- dating, andflexibility are developed forbotha classofcurrent sequential systemsand an associative system.Utilizing these models, a comparison ismadebetweenthetwosystemsunder certain assumptions sothatforgivenclasses ofdatamanagement problems, one candetermine whichsystemismore favorable. Thecomparison reveals that, ingeneral, theassociative system outperforms current inverted list systems thatuse pointers. Inthe areaofretrieval, itisshownthatforsingle criteria searches the performance improvement isproportional tothelogarithm ofthe list length. Formulticriteria searches involving theunion orinter- section ofitems, a significantly higher performance improvement is obtained. Forexample, ratios of50to1 arecommon. Itisshownthat current systems require atleast twice asmuchstorage astheassocia- tive system. Also, updating may besignificantly faster. Forexample, toupdate oneiteminalist of16,theratio ofinterrogations isabout 30to1.However, whenallitems inthelist areupdated, theratio is onlyabout 3to1.Flexibility isa measure ofhoweasily a data man- agementsystemresponds tomodified requirements suchas new queryclasses, different datastructures, etc.Itisshownthatthe associative systemcan easily adapttomodified requirements, thereby providing a highdegree offlexibility, whereas incurrent sequential systems, flexibility isrelatively poor.Forexample, inthe associative system, indexing ispotentially available atthebitlevel without a storage penalty. On theotherhand,incurrent systems indexing requires ahighdegree ofdataredundancy. IndexTerms-Associative memories, content-addressable mem- ories, datamanagement,datastructures, information systems, inverted lists.