Verifiable auctions for ad exchanges

This paper is about a critical component of the Web ecosystem that has so far received little attention in our community: ad exchanges. They run auctions to sell publishers’ inventory—space on a Web page—to advertisers who want to display ads in those spaces. Unfortunately, under the status quo, the parties to the auction cannot check that an auction was carried out correctly, which raises the following more general question: how can we create verifiability in low-latency, high-frequency auctions where the parties do not know each other? We address this question with the design, prototype implementation, and experimental evaluation of VEX. VEX introduces a technique for efficient, privacypreserving integer comparisons, couples these with careful protocol design, and guarantees that an incorrectly computed auction is detectable by all parties. We find that VEX adds little latency and significant (but tolerable) processing cost, and is practical, or nearly so.