CAREER: Introspective Computing: A Multicore Approach to Availability, Reliability and Security

Georgia Tech Research Corporation, Atlanta GA

Investigators

Abstract

Hsien-Hsin S Lee Georgia Institute of Technology CAREER: Introspective Computing: A Multcore Approach to Availability, Reliability, and Security 0644096 Panel ID: 070111 Abstract Cyber exploits and remote attacks on network servers are becoming enormous concerns by service providers, online merchants, and network users. Due to the deficiency of effective and efficient countermeasures and self-healing capability, malicious adversaries can easily compromise systems, gain unauthorized access, and disrupt services, leading to huge loss of productivity and revenues. To provide high availability, reliability and security for computing systems is, more urgent than ever, the first-order design criteria. On the other hand, technology scaling has reached a point where processor architects can integrate multiple processor cores onto the same die. Using such multi-core processors with necessary hardware and system software support, an introspective computing platform can be constructed to addresses these emerging issues. An introspective multi-core architecture leverages several advantages offered by the tightly-coupled on-die processor cores to perform: fine-grained security introspection, instant low-overhead checkpoint, and fast, on-demand rollback recovery, thereby providing a continuing service in the face of system compromise or corruption. The key features of an introspective system include programmable asymmetric processor cores, physical insulation against exploits, highly efficient fine-grained introspection, and fast checkpoint/recovery mechanism. In addition, with adequate monitoring and detection mechanism implemented, fault-tolerance and hardware-based transactional memory system can be enabled for minimal cost. Overall, it provides a synergistic and holistic solution toward the challenges of achieving high availability, reliability, and security. The outcome of this research will substantially benefit the IT industry, the generic users for protecting their privacy, and the processor architects to consider these challenges in the early design phase.