Processing numerical data securely in NoSQL database backed shared user spaces

Abstract

Performingvariousmathematicaloperations,searchingandsortingonacloudenvi- ronmentaresecurityconcerns.Sincetheuserspaceissharedtherearequestionsabout con dentialityofthedata.Byperformingoperationsonencrypteddata,con dentiality can beachievedwhiletheyarebeingprocessed.Thiscanbedoneusinghomomorphic encryption. Alsoorderpreservingencryptionschemescanbeusedtoprovidesearching and sortingfortheencrypteddata. When relatedworkisconsidered,someoftheimplementationslikecipherbaseand trustdb requirespecialhardwaresetup,somearehighlycomplexandimplementations likecryptdbrequiresmallchangestothedatabasecoreaswell.Thusnonecanbeused in acloudenvironmentunlesstheenvironmentbelongstothesamepersonorcompany. This endeavoristoprovideanopensourceimplementationthatworksinlooselycoupled manner toprovideabovecapabilitiesfordatastoredinCassandra. As anoutcomeoftheresearchanopensourcelibraryisimplementedwithhomomor- phic encryptioncapabilitieswhichcanperformmathematicaloperationsandsearchand sort onencrypteddata.Basicmathematicaloperationsaresupported(i.e.addition,sub- traction, multiplication,division)intheimplementationforintegers,realnumbersand negativevalues,usingPaillierandRSAciphers.Tomakethedataatrestsemantically secure asecondlayerofencryptionisadded.Anorderpreservingencryptionscheme called mOPEisusedtoprovidesearchingandsorting.Cassandraisusedasthedatabase to persistthedatasincelotofcloudserviceproviders'useitintheirinfrastructures. Based ontheevaluationtheimplementedlibraryperformsconsistentlyformathemat- ical operationsregardlessoftheplaintextsizeencrypted.Alsobasedonthenumberof operationstheforcipher-textaddition,thetimegrowsexponentially.Encryptionislinear and theperformanceisnotchangedbasedonthekeysizeofthecipher-text.Whenrecords with encryptedvaluesareinsertedintoCassandra,performancedroppedapproximately 8 times. As futureworkperformanceimprovementstotheimplementation,performingcomplex functions oncipher-text,completesearchingandsortingonencryptedtextcanbedone. Evaluationssuggestthateventhoughthisresearchisstillnotreadytobeintegrated with arealtimesystem,itcanbeusedinscenarioslikebatchprocessingordatamining independentlyoftheinfrastructureused.