The International Conference on Dependable Systems and Networks (DSN-2006)

Reliability-Aware Microprocessor Architectures

Sarita Adve and Pradip Bose, University of Illinois and IBM TJ Watson Research

• Click for detailed description.
  In this tutorial, we present the foundational principles and methodologies behind the design of microprocessors that meet market-driven reliability targets, in the face of technological constraints offered by the late CMOS era. The stress is on early-stage (pre-RTL) definition at the microarchitecture level, although relevant details from lower levels of design (e.g. logic, circuits and below) are also covered where appropriate. In particular, in order to explain the methodology for modeling the effects of failures at the microarchitectural level, we delve into some details of the actual physical failure mechanisms, and the models that govern their onset and propagation.

  We first cover the topic of pre-silicon modeling to estimate performance, power, temperature and reliability, in the context of target workloads of interest to the design team. While estimating failure rates and mean time to failure (MTTF), we consider the effects of hard (or permanent) failures, as well as soft (or transient) errors. We address the issues of modeling accuracy and validation in some detail. In particular, we examine the basic axioms that guide alternate modeling methodologies, and discuss the range of validity of these assumptions. We also touch on the important topic of reliability "metrics": we discuss the issue of defining appropriate metrics to quantify a given index of reliability, and the need to guard against fallacies and pitfalls when trying to interpret a projected reliability value. Subsequently, we cover the topic of reliability-aware design at the microarchitectural level. We discuss power-, area- and performance-efficient approaches to provide temporal and/or spatial redundancy support in order to meet required reliability targets. We also discuss adaptive microarchitectures: those that are designed to change with variations in the workload, with the goal of maximizing one or more of: reliability, power-temperature efficiency and performance. In discussing each topic area within the tutorial, we provide a brief survey of past techniques and results, before providing in-depth coverage of more recently published methodologies.

Dependable Computing over Sensor Networks

Shivkant Mishra, University of Colorado

• Click for detailed description.
  Wireless sensor networks are rapidly growing in their importance and relevance to both the research community and the public at large. In a wireless sensor network, a distributed collection of sensor nodes forms a network interconnected by wireless communication links. Each sensor node acts as information source, sensing and colleting data samples from its environment. Sensor nodes perform routing functions, creating a multi-hop wireless networking fabric that relays data samples to other sensor nodes and to external destinations. Applications of wireless sensor networks are numerous, diverse, and growing. They range from habitat monitoring to indoor monitoring of semiconductor fabrication processes, and from counter-sniper localization on battlefields to search and rescue operations. In each of these application scenarios, lives and livelihoods may depend on the timeliness and correctness of the sensor data obtained from dispersed sensor nodes.

  The main objective of this tutorial is to provide an in-depth coverage of design and implementation issues in building a dependable wireless sensor network, and cover the current state-of-the-art of this promising technology. The tutorial will first provide a basic introduction to the wireless sensor networks, and then cover five important technical issues in the design and implementation of a dependable wireless sensor network. These five issues are cryptographic key management, secure and intrusion-tolerant routing, secure in-network processing, secure dynamic reprogramming, and prevention against traffic analysis attacks.

  Because of the importance and sensitivity of tasks performed by wireless sensor networks, a wireless sensor network is a target of adversaries. There are many security threats to wireless sensor networks. An adversary can prevent users from getting correct data from sensor nodes by modifying the contents of the packets, or spoofing the identity of the sensor nodes. An adversary can block communication between a base station and sensor nodes by creating false routing information, or simply generating jamming signals. An adversary can gain control of an entire sensor network by spoofing the identity of a base station. An adversary can compromise a sensor node, get all information from that node, and can even re-program it to behave like a malicious node.

  The design and implementation of a dependable wireless sensor network must simultaneously address three research challenges: (1) Resource constraints of sensor nodes in terms of low power, low memory, slower CPU, and limited communication bandwidth; (2) Vulnerability of wireless communication to eavesdropping, unauthorized access, spoofing, replay, and denial of service attacks; and (3) Added physical security risk of individual sensor nodes falling into the wrong hand and be compromised.

  This tutorial will discuss some of the latest techniques that have been proposed to address these research challenges. Finally, a case study of a search-and-rescue application built using a wireless sensor network will be done.

More Reliable Software Faster and Cheaper

John Musa, Consultant

• Click for detailed description.
  Stressed out by competitive pressures to deliver more reliable software faster and cheaper? Want to control the process rather than have it control you? Software reliability engineering (SRE), a practice primarily developed on the job in industry, can help. This unique tutorial will teach you the essentials of how to apply it.

  SRE is based on two powerful ideas:

  • Quantitatively characterize expected use and then focus resources on most used and/or most critical functions. This increases development efficiency and hence effective resource pool available to add customer value to product.
  • Further increase customer value by setting quantitative reliability objectives that precisely balance customer needs for reliability, timely delivery, and cost; engineer project strategies to meet them; and track reliability in test as a release criterion

  SRE is a standard, proven best practice. It applies not only to software reliability but to dependability and safety as well. You can apply it to any system using software and to members of software component libraries. And you can start with the next release.

Architecture Level Evaluation of Soft Errors

Shubu Mukherjee, Intel Corp.

• Click for detailed description.
  Tutorial contents:
  1. The Soft Error problem & Motivation (45 minutes)
  2. AVF (arch. vulnerability factor) Basics (30 minutes)
  3. Computing AVF using Statistical Fault Injection (15 minutes)
  4. Computing AVF using ACE & lifetime analysis (30 minutes)
  5. Computing AVF of Address-Based Structures (30 minutes)
  6. Examples & Results (15 minutes)
  7. AVF Reduction Techniques (30 minutes)
  8. Future use of AVF techniques (15 minutes)
  9. Open discussion (15 minutes)

Erasure Codes for Fault Tolerant Storage

James S. Plank, University of Tennessee

• Click for detailed description.
  Tutorial contents:
  • To introduce you to the various erasure coding techniques.
    • Reed Solomon codes.
    • Parity-array codes.
    • LDPC codes.
  • To help you understand their tradeoffs.
  • To help you evaluate your coding needs.
    • This too is not straightforward.

Software Dependability: What You Didn't Learn in Kindergarten

John C. Knight and Elisabeth Strunk, University of Virginia

• Click for detailed description.
  This tutorial will discuss a number of important topics in software system dependability with a heavy emphasis on their practical relevance. The focus will be an example application domain, digital avionics. Starting with a brief summary of that domain, the tutorial will explore several topics including: (1) Basic principles of software dependability; (2) the practical use of formal specification in real systems using Z and VDM; (3) the practical use of formal verification; (4) static analysis and the associated role of the programming language; (5) the assessment of software systems by statistical means and safety cases; and (6) the role of standards.

  This tutorial will be of interest to engineers and managers engaged in the development of software systems for high-assurance applications in all domains. No prior technical knowledge is assumed beyond what the attendee learned in kindergarten.