NeTS: CSR: Small: Towards a Redundancy Aware Network Stack

Tufts University, Medford MA

Investigators

Abstract

This research project will leverage redundancy, which is naturally present in typical cloud environments, to improve application performance. The key idea is to make the entire network stack aware of replicas, and to design network mechanisms that pro-actively use duplicate requests to avoid stragglers. Specifically, this research will build a redundancy aware network stack (RANS); in RANS, applications will expose information about redundancy to the network (e.g., data is stored in both machine A and machine B). The network will use new mechanisms that leverage this information (e.g., transport protocol that uses multiple replicas at the same time). Applications will be modified to make full use of the new mechanisms. By exploring the co-design of the application and the network, this project will make research advances at different levels of the protocol stack. This includes innovation in at least four different aspects: i) scheduling framework, ii) abstraction and interface between the applications and the network, iii) new network mechanisms (transport, in-network mechanisms), and iv) new application designs that exploit the new mechanisms. This research will address the challenges of implementing this framework for different resources (e.g., network). RANS promises to overcome the problems caused by stragglers. This will reduce application latency, especially at the tail, which is a major concern for application providers. The proposed research will make fundamental advances in designing systems and network protocols to improve performance of data center applications. This research will have significant impact on various stakeholders, including Internet users, cloud service providers, and researchers and students of networking. End users will experience better performance while using existing applications. Cloud service providers can deploy new applications that require high reliability and performance. Researchers will benefit from the open source software artifacts and students will benefit from a cloud computing Hackathon workshop and opportunities for research projects.