GPU-based Multi-stream Analyzer on Application Layer for Service-Oriented Router

Abstract

A service-oriented router (SoR) is a new router architecture that provides rich application-layer services to Internet users by extracting information from network traffic. SoR performs stream reconstruction and selection using string matching. After on-the-fly reconstruction of stream data, SoR extracts the required information using a software-based character string analyzer. For wire-rate string matching and for design flexibility, a software-based string-matching accelerator is required. A graphics processing unit (GPU) is a promising solution as a conventional cost-effective accelerator for use in a high-end router. Existing string-matching methods do not consider the multiple-stream processing and effective thread management required in a network application. We propose an accelerated string-matching method that is customized for a router and implemented a task controller that improves thread distribution depending on the status of core processes.