Collaborative Path Prediction in Cache Pre-fetching for Distributed State Machines

Abstract

Modeling distributed processes using state machines is gaining importance as the serverless computing becomes more widespread. One of the important issues in distributed state machine execution is to perform better caching approaches. Due to the nature of state machines, the amount of different paths that can be executed by a state machine is limited. This property can be utilized to perform better caching by predicting which path is going to be taken. In this paper, we analyze former execution history of distributed state machines to perform better predictions. We also run experiments to find out if collaboratively using the history of the state machine instances makes any significant improvements on pre-fetching decision. Our results show that pre-fetching significantly decrease the number of cache misses while history sharing between instances provides improvements in a more limited fashion.