SafetyNet: Secure communications for embedded high-performance computing

ISI's Advanced Scalable Network Technology (ASNT) project employs multiple specialized network components in an integrated high-performance multicomputer system. Large multicomputer systems are vulnerable to programming error, hardware faults, and potentially, malicious attacks. Allowing direct user access to network interfaces is desirable for performance, but can reduce protection and reliability. Even privileged codes, such as general-purpose operating systems, are untrustworthy and buggy, ASNT's security and reliability subsystems, collectively known as SafetyNet, embed vital distributed-system communications and authentication functions to prevent even OS code from breaching security or bringing the system down. Such mechanisms are necessary to enable deployment of embedded HPC systems in mission-critical applications. In this paper we discuss the motivation and features of the SafetyNet system for embedded high-performance computing.