Fault Tolerant QR Factorization for General Matrices

TL;DR

This paper introduces a fault-tolerant QR factorization algorithm for general matrices that uses communication-avoiding techniques and redundancy to recover from process failures without significant overhead.

Contribution

It proposes a novel fault-tolerant QR algorithm that enables process failure recovery with minimal additional computation and no impact on normal execution.

Findings

Enables recovery of failed process state from a single process.

Maintains efficiency during failure-free execution.

Uses structure of reduction to introduce redundancies.

Abstract

This paper presents a fault-tolerant algorithm for the QR factorization of general matrices. It relies on the communication-avoiding algorithm, and uses the structure of the reduction of each part of the computation to introduce redundancies that are sufficient to recover the state of a failed process. After a process has failed, its state can be recovered based on the data held by one process only. Besides, it does not add any significant operation in the critical path during failure-free execution.