How to Make Replicated Data Secure

Many distributed systems manage some form of long-lived data, such as files or data bases. The performance and fault-tolerance of such systems may be enhanced if the repositories for the data are physically distributed. Nevertheless, distribution makes security more difficult, since it may be difficult to ensure that each repository is physically secure, particularly if the number of repositories is large. This paper proposes new techniques for ensuring the security of long-lived, physically distributed data. These techniques adapt replication protocols for fault-tolerance to the more demanding requirements of security. For a given threshold value, one set of protocols ensures that an adversary cannot ascertain the state of a data object by observing the contents of fewer than a threshold of repositories. These protocols are cheap; the message traffic needed to tolerate a given number of compromised repositories is only slightly more than the message traffic needed to tolerate the same number of failures. A second set of protocols ensures that an object’s state cannot be altered by an adversary who can modify the contents of fewer than a threshold of repositories. These protocols are more expensive; to tolerate t-1 compromised repositories, clients executing certain operations must communicate with t-1 additional sites. This research was sponsored by the Defense Advanced Research Projects Agency (DOD), ARPA Order Numbers 4864 and 4976, monitored by the Air Force Avionics Laboratory under Contracts F33615-84K-1520 and N00039-84-C-0467.