Off-campus UMass Amherst users: To download campus access dissertations, please use the following link to log into our proxy server with your UMass Amherst user name and password.

Non-UMass Amherst users: Please talk to your librarian about requesting this dissertation through interlibrary loan.

Dissertations that have an embargo placed on them will not be available to anyone until the embargo expires.

Document Type

Open Access Dissertation

Degree Name

Doctor of Philosophy (PhD)

Degree Program

Electrical and Computer Engineering

Year Degree Awarded

2015

Month Degree Awarded

September

First Advisor

Russell Tessier

Second Advisor

Tilman Wolf

Subject Categories

VLSI and Circuits, Embedded and Hardware Systems

Abstract

As an essential part of modern society, the Internet has fundamentally changed our lives during the last decade. Novel applications and technologies, such as online shopping, social networking, cloud computing, mobile networking, etc, have sprung up at an astonishing pace. These technologies not only influence modern life styles but also impact Internet infrastructure. Numerous new network applications and services require better programmability and flexibility for network devices, such as routers and switches. Since traditional fixed function network routers based on application specific integrated circuits (ASICs) have difficulty keeping pace with the growing demands of next-generation Internet applications, there is an ongoing shift in the industry toward implementing network devices using programmable network processors (NPs).

While network processors offer great benefits in terms of flexibility, their reprogrammable nature exposes potential security risks. Similar to network end-systems, such as general-purpose computers, software-based network processors have security vulnerabilities that can be attacked remotely. Recent research has shown that a new type of data plane attack is able to modify the functionality of a network processor and cause a denial-of-service (DoS) attack by sending a single malformed UDP packet. Since this attack relies solely on data plane access and does not need access to the control plane, it can be particularly difficult to control.

Hardware security monitors have been introduced to identify and eliminate these malicious packets before they can propagate and cause devastating effects in the network. However, previous work on hardware monitors only focus on single core systems with static (or very slowly changing) workloads. In network processors that use up to hundreds of parallel processor cores and have processing workloads that can change dynamically based on the network traffic, the realization of a complete multicore hardware monitoring system remains a critical challenge. Our research work in this thesis provides a comprehensive solution to this problem.

Our first contribution is the design and prototype implementation of a Scalable Hardware Monitoring Grid (SHMG). This scalable architecture balances area cost and performance overhead by using a clustered approach for multicore NP systems. In order to adapt to dynamically changing network traffic, a resource reallocation algorithm is designed to reassign the processing resources in SHMG to different network applications at runtime. An evaluation of the prototype SHMG on an Altera DE4 board demonstrates low resource and performance overheads. The functionality and performance of a runtime resource reallocation algorithm are tested using a simulation environment.

A second significant contribution of this work is a network system-level security solution for multicore network processors with hardware monitors. It addresses two key problems: (1) how to securely manage and reprogram processor cores and monitors in a deployed router in the network, and (2) how to prevent the large number of identical router devices in the network from an attack that can circumvent one specific monitoring system and lead to Internet-scale failures. A Secure Dynamic Multicore Hardware Monitoring System (SDMMon) is designed based on cryptographic principles and suitable key management to ensure the secure installation of processor binaries and monitor graphs. We present a Merkle tree based parameterizable high performance hash function that can be configured to perform a variety of functions in different devices via a 32-bit configuration parameter. A prototype system composed of both the SDMMon and the parameterizable hash is implemented and evaluated on an Altera DE4 board.

Finally, a fully-functional, comprehensive Multicore NP Security Platform, which integrates both the SHMG and the SDMMon security features, has been implemented on an Altera DE5 board.

Share

COinS