Efficient handling of lots of simulation data files

Abstract

Saving information to files is the most basic and simplest way to store data, so it is often used in simple simulators and simulation tools as the first choice for logging information about the simulation process and its results. Computer simulations often involve simulating a significant number of replications and accumulating large numbers of files. Today's filesystems are still not capable of efficiently storing and processing millions of files. This paper presents alternatives that allow for more efficient storage, transfer, and analysis of data, with an emphasis on easy migration or implementation from the initial data files. Simple approaches such as using TAR or ZIP archives to sophisticated approaches involving Parquet file, S3-like object storage (e.g., MinIO, OpenIO) and (non)relational databases will be compared.