Efficient and Robust Distributed Matrix Computations via Convolutional Coding

Distributed matrix computations - matrix-matrix or matrix-vector multiplications are well-recognized to suffer from the problem of stragglers (slow failed worker nodes). Much prior work in this area is (i) either sub-optimal terms its straggler resilience, (ii) suffers numerical problems, i.e., there a blow-up round-off errors decoded result owing high condition numbers corresponding decoding matrices. Our presents convolutional coding approach that removes these limitations. It optimal and has excellent robustness as long workers' storage capacity slightly higher than fundamental lower bound. Moreover, it can be using fast peeling decoder only involves add/subtract operations. second marginally complexity first one, but allows us operate arbitrarily close Its theoretically quantified by deriving computable upper bound on worst case number over all possible matrices drawing connections with properties large block Toeplitz All above claims backed up extensive experiments done AWS cloud platform.