Version 1. 1 (Revised and expanded) This version introduces several improvements to the original DME paper: - Reformulated Diversity Divergence Theorem with explicit three-part structure and probabilistic bound P ≤ (ρL/R) ^N-1- Added Assumption 2. 2 (Replica-Private Address Spaces) to strengthen fault model- Clarified memory footprint: 512 bytes per replica covers CPU context + hash state only (stack/heap allocated separately) - Added Listing 1 with per-replica execution loop showing fetch-execute-hash-compare cycle- Added note that while tested in virtualised environment, DME can be implemented natively in assembly on physical hardware- Minor corrections and formatting improvements Original abstract: Redundancy-based fault tolerance techniques typically execute identical binaries with identical address layouts, leaving systems vulnerable to correlated control-flow faults. This paper introduces Divergent Multi-Version Execution (DME), which combines address-space decorrelation with per-instruction full-state hashing. Identical instruction bytes are preserved across replicas, while basic blocks are mapped to distinct addresses. After each instruction, replicas compute incremental state hashes and perform synchronous comparison.
Petro Baran (Wed,) studied this question.