Intelligent Software-Defined Storage With Deep Traffic Modeling for Cloud Storage Service

Abstract

The advent of cloud computing, big data, and mobile computing has created a fast-growing demand for storage. Cloud service providers are looking for cost-effective storage solutions as an alternative to traditional, high-cost, embeddedsystems-based storage to meet the needs of newly emerging applications, such as messaging, video streaming, data analytics, etc. In particular, they are facing the challenge of lowering costs while still accommodating multi-workloads on a single instance of storage without compromising workload performance requirements. Software-defined storage (SDS) is a new generation of storage system. Unlike traditional embedded-systems-based storage, the SDS uses a software-stack above commodity hardware to provide more valuable and cost-effective features. To meet the challenges cloud service providers are facing, this paper introduces the architecture of a new SDS platform called Federator. It also argues that the architecture of an SDS platform should have three main characteristics: 1. separation of the control and data pathways, 2. self-configuration of storage resources, and 3. RESTful APIs for new business extension. This paper specifically introduces the storage I/O traffic modeling supported by Federator. With this capability, storage performance metrics are generated by using Long-Short Term Memory (LSTM). This prediction capability is important to a self-configurable SDS to meet performance requirements.