Terviklik protokollide kogu kahe osapoolega turvalisteks arvutusteks
Date
2016
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
Turvaline ühisarvutus võimaldab üksteist mitte usaldavatel osapooltel teha arvutusi tundlikel andmetel nii, et kellegi privaatsed andmed ei leki teistele osapooltele. Sharemind on kaua arenduses olnud turvalise ühisarvutuse platvorm, mis jagab tundlikke andmeid ühissalastuse abil kolme serveri vahel. Sharemindi kolme osapoolega protokolle on kasutatud suuremahuliste rakenduste loomisel. Igapäevaelus leidub rakendusi, mille puhul kahe osapoolega juurustusmudel on kolme osapoolega variandist sobivam majanduslikel või organisatoorsetel põhjustel. Selles töös kirjeldame ja teostame täieliku protokollistiku kahe osapoolega turvaliste arvutuste jaoks. Loodud protokollistiku eesmärk on pakkuda kolme osapoolega juurutusmudelile võrdväärne alternatiiv, mis on ka jõudluses võrreldaval tasemel. Kahe osapoole vahelised turvalise aritmeetika protokollid tuginevad peamiselt Beaveri kolmikute ette arvutamisele. Selleks, et saavutada vajalikku jõudlust, oleme välja töötanud tõhusad ette arvutamise meetodid, mis kasutavad uudsel viisil N-sõnumi pimeedastuse pikendamise protokolle. Meie meetodite eeliseks on alternatiividest väiksem võrgusuhtluse maht. Töös käsitleme ka insenertehnilisi väljakutseid, mis selliste meetodite teostamisel ette tulid. Töös esitame kirjeldatud konstruktsioonide turvalisuse ja korrektsuse tõestused. Selleks kasutame vähem eelduseid, kui tüüpilised teaduskirjanduses leiduvad tõestused. Üheks peamiseks saavutuseks on juhusliku oraakli mudeli vätimine. Meie kirjeldatud ja teostatud täisarvuaritmeetika ja andmetüüpide vaheliste teisendusprotokollide jõudlustulemused on võrreldavad kolme osapoole protokollide jõudlusega. Meie töö tulemusena saab Sharemindi platvormil teostada kahe osapoolega turvalisi ühisarvutusi.
Secure multi-party computation allows a number of distrusting parties to collaborate in extracting new knowledge from their joint private data, without any party learning the other participants' secrets in the process. The efficient and mature Sharemind secure computation platform has relied on a three-party suite of protocols based on secret sharing for supporting large real-world applications. However, in some scenarios, a two-party model is a better fit when no natural third party is involved in the application. In this work, we design and implement a full protocol suite for two-party computations on Sharemind, providing an alternative and viable solution in such cases. We aim foremost for efficiency that is on par with the existing three-party protocols. To this end, we introduce more efficient techniques for the precomputation of Beaver triples using oblivious transfer extension, as the two-party protocols for arithmetic fundamentally rely on efficient triple generation. We reduce communication costs compared to existing methods by using 1-out-of-N oblivious transfer extension in a novel way, and provide insights into engineering challenges for efficiently implementing these methods. Furthermore, we show security of our constructions using strictly weaker assumptions than have been previously required by avoiding the random oracle model. We describe and implement a large amount of integer operations and data conversion protocols that are competitive with the existing three-party protocols, providing an overall solid foundation for two-party computations on Sharemind.
Secure multi-party computation allows a number of distrusting parties to collaborate in extracting new knowledge from their joint private data, without any party learning the other participants' secrets in the process. The efficient and mature Sharemind secure computation platform has relied on a three-party suite of protocols based on secret sharing for supporting large real-world applications. However, in some scenarios, a two-party model is a better fit when no natural third party is involved in the application. In this work, we design and implement a full protocol suite for two-party computations on Sharemind, providing an alternative and viable solution in such cases. We aim foremost for efficiency that is on par with the existing three-party protocols. To this end, we introduce more efficient techniques for the precomputation of Beaver triples using oblivious transfer extension, as the two-party protocols for arithmetic fundamentally rely on efficient triple generation. We reduce communication costs compared to existing methods by using 1-out-of-N oblivious transfer extension in a novel way, and provide insights into engineering challenges for efficiently implementing these methods. Furthermore, we show security of our constructions using strictly weaker assumptions than have been previously required by avoiding the random oracle model. We describe and implement a large amount of integer operations and data conversion protocols that are competitive with the existing three-party protocols, providing an overall solid foundation for two-party computations on Sharemind.