Learning-Aided Heuristics Design for Storage System

Abstract

Computer systems such as storage systems normally require transparent white-box algorithms that are interpretable for human experts. In this work, we propose a learning-aided heuristic design method, which automatically generates human-readable strategies from Deep Reinforcement Learning (DRL) agents. This method benefits from the power of deep learning but avoids the shortcoming of its black-box property. Besides the white-box advantage, experiments in our storage production's resource allocation scenario also show that this solution outperforms the system's default settings and the elaborately handcrafted strategy by human experts. CCS CONCEPTS• Computing methodologies → Planning with abstraction and generalization; Rule learning; • Computer systems organization → Real-time operating systems.

LQoCo: Learning to Optimize Cache Capacity Overloading in Storage Systems

“…The workloads fall into two classes, i.e., standard workload traces and real-world traces from the service of customers. For the standard benchmark workload traces, the open-source benchmarks including Filebench, FIO, and Vdbench are utilized, each of which is associated with one typical business model of the users, such as database, heavy computing in AI tasks or high performance computing [14]. These tools are commonly used to test and validate storage systems.…”

Building efficient and practical machine learning systems

“…Bao [283] is a learning-based system that adopts TreeCNN [333] for query optimization and the decision process can be inspected by developers. Tang et al [334] proposed an interpretable method that extracts a Finite State Machine from a RL policy for storage resource allocation in Huawei. Grüner et al [335] generated concise and interpretable rule-sets for unknown proprietary streaming algorithms (e.g.…”