Computer System Modeling

Including Storage, Processor, Memory, etc. (Published in IEEE CAL'24, ISPASS'26)

This is another important theme of my research in the past 10 years, and is still an important theme.

Computer systems are complex. To design a new architecture, typically implementing in the real system is not the first step. To allow for design space explorations and also minimize the design costs, modeling the real system in a simulator or emulator or even a simple regression model is the true first step. In this regard, we have always paid special attention to system characterization and behavior modeling. An example is our IEEE CAL’24 in which we characterize a storage system with 72 SSDs and model how adding SSDs and disk enclosures affect system performance. Another recent work that we have has been published in ISPASS’26. We model the performance storage systems using machine learning models and propose a framework to more accurately report True Utilization of each storage device.

Our proposed framework to accurate calculate True Utilization of a storage device. We partly use machine learning models in our framework (from our ISPASS'26 paper).