Characterization of I / O Patterns generated by Fault Tolerance in HPC environments

Fault Tolerance (FT) is an important strategy to guarantee the availability of systems in High-Performance Computing environments (HPC), and for this purpose it offers several models that allow the recovery and reconfiguration of systems in the event of failures. This protection involves the use of protocols that require continuous, simultaneous and large-scale access to stable storage through Input and Output (I/O) operations, which is one of the main causes of the overhead generated by the protection of FT. In this paper, we propose a methodology to characterize the behavior of these I/O patterns that generates fault tolerance, as well as to evaluate the amount of information that must be stored, since the management of stable storage affects the overhead of the fault tolerance scheme to a large extent. Going deeper into these elements will allow us to better manage the volume of information generated and select the appropriate storage system.