Periodic I/O scheduling for super-computers

Abstract

With the ever-growing need of data in HPC applications, the congestion at the I/O level becomes critical in super-computers. Architectural enhancement such as burst-buffers and pre-fetching are added to machines, but are not sufficient to prevent congestion. Recent online I/O scheduling strategies have been put in place, but they add an additional congestion point and overheads in the computation of applications. In this work, we show how to take advantage of the periodic nature of HPC applications in order to develop efficient periodic scheduling strategies for their I/O transfers. Our strategy computes once during the job scheduling phase a pattern where it defines the I/O behavior for each application, after which the applications run independently, transferring their I/O at the specified times. Our strategy limits the amount of I/O congestion at the I/O node level and can be easily integrated into current job schedulers. We validate this model through extensive simulations and experiments by comparing it to state-of-the-art online solutions. Specifically, we show that not only our scheduler has the advantage of being de-centralized, thus overcoming the overhead of online schedulers, but we also show that on Mira one can expect an average dilation improvement of 22% with an average throughput improvement of 32%! Finally, we show that one can expect those improvements to get better in the next generation of platforms where the compute - I/O bandwidth imbalance increases.