Modeling Compute Storage Quality of Service and Latency using Sigmoid Functions

Due to the growing demand of data platforms, popularity of the solid-state drive (SSD) has risen to boost performance, reliability, and endurance and to minimize power consumption. Consequently, performance modeling of SSD systems has sought to give insights into current and novel storage architectures. To date, these models have fallen short of real-world applicability; their limited scope fail to accurately account for the stochastic behavior of I/O storage request latencies. Therefore, developing a method to obtain accurate, variable, and probabilistic latencies is imperative to the future success of these models. In this paper, we propose a method for modeling the SSD from empirical performance data to predict key performance indicators (KPIs). Using data from industry standard Flexible I/O (FIO) testing, we derive a computable distribution of latencies. Comparing the results of our model to the test data, we found that we had a lower percentage error than other simulation and performance modeling solutions. Additionally, we show that our model can be used to treat various classes of SSDs under diverse workloads and discuss how it can be applied to inform design of computer systems.