Locally Recoverable Coded Matrix Multiplication
Haewon Jeong, Fangwei Ye, Pulkit Grover · 2018
Repair locality is important to recover from failed nodes in distributed computing especially when communicating all the data to a master node is expensive. Here, building on recent work on coded matrix multiplication, we provide locally recoverable coded matrix multiplication strategies. Leveraging constructions of optimal matrix multiplication codes and optimal locally recoverable (LRC) codes, we provide constructions of LRC Polynomial codes (minimal communication) and LRC MatDot codes (minimal storage).