Professor Snir is interested in various aspects of parallel computing and large-scale distributed processing. His theoretical work in this area includes studies of parallel computation models and of parallel algorithms. His work in parallel computer architecture includes work on parallel interconnection networks and on mechanisms for efficient large-scale shared memory support. His work on parallel computing software includes contributions to the design and implementation of the MPI message-passing library, work on performance visualization tools, and work on a parallel file system. While at IBM Research he contributed to the design of the IBM SP2 supercomputer system and several large-scale shared memory systems.
More recently Prof. Snir has worked on reliability issues in large-scale parallel systems. These includes mechanisms for predicting impeding failures, detecting silent errors, and improved checkpoint-recovery algorithms.