Contents

• 1. Introduction
  • 1.1 Purpose
  • 1.2 History and Motivation
  • 1.3 Synthesis Overview
    • 1.3.1 Kernel Structure
    • 1.3.2 Implementation Ideas
    • 1.3.3 Implementation Language
    • 1.3.4 Target Hardware
    • 1.3.5 Unix Emulator
• 2. Previous Work
  • 2.1 Overview
  • 2.2 The Tradeoff Between Throughput and Latency
  • 2.3 Kernel Structure
    • 2.3.1 The Trend from Monolithic to Diffuse
    • 2.3.2 Services and Interfaces
    • 2.3.3 Managing Diverse Types of I/O
    • 2.3.4 Managing Processes
• 3. Kernel Code Generator
  • 3.1 Fundamentals
  • 3.2 Methods of Runtime Code Generation
    • 3.2.1 Factoring Invariants
    • 3.2.2 Collapsing Layers
    • 3.2.3 Executable Data Structures
    • 3.2.4 Performance Gains
  • 3.3 Uses of Code Synthesis in the Kernel
    • 3.3.1 Buffers and Queues
    • 3.3.2 Context Switches
    • 3.3.3 Interrupt Handling
    • 3.3.4 System Calls
  • 3.4 Other Issues 3.4.1 Kernel Size
    • 3.4.2 Protecting Synthesized Code
    • 3.4.3 Non-coherent Instruction Cache
  • 3.5 Summary
• 4. Kernel Structure
  • 4.1 Quajects
    • 4.1.1 Quaject Interfaces
    • 4.1.2 Creating and Destroying Quajects
    • 4.1.3 Resolving References
    • 4.1.4 Building Services
    • 4.1.5 Summary
  • 4.2 Procedure-Based Kernel
    • 4.2.1 Calling Kernel Procedures
    • 4.2.2 Protection
    • 4.2.3 Dynamic Linking
  • 4.3 Threads of Execution
    • 4.3.1 Execution Modes
    • 4.3.2 Thread Operations
    • 4.3.3 Scheduling
  • 4.4 Input and Output
    • 4.4.1 Producer/Consumer
    • 4.4.2 Hardware Devices
  • 4.5 Virtual Memory
  • 4.6 Summary
• 5. Concurrency and Synchronization
  • 5.1 Synchronization in OS Kernels
    • 5.1.1 Disabling Interrupts
    • 5.1.2 Locking Synchronization Methods
    • 5.1.3 Lock-Free Synchronization Methods
    • 5.1.4 Synthesis Approach
  • 5.2 Lock-Free Quajects
    • 5.2.1 Simple Linked Lists
    • 5.2.2 Stacks and Queues
    • 5.2.3 General Linked Lists
    • 5.2.4 Lock-Free Synchronization Overhead
  • 5.3 Threads
    • 5.3.1 Scheduling and Dispatching
    • 5.3.2 Thread Operations
    • 5.3.3 Cost of Thread Operations
  • 5.4 Summary
• 6. Fine-Grain Scheduling
  • 6.1 Scheduling Policies and Mechanisms
    • 6.2.1 Hardware Phase Locked Loop
    • 6.2.2 Software Feedback
    • 6.2.3 FLL Example
    • 6.2.4 Application Domains
  • 6.3 Uses of Feedback in Synthesis
    • 6.3.1 Real-Time Signal Processing
    • 6.3.2 Rhythm Tracking and The Automatic Drummer
    • 6.3.3 Digital Oversampling Filter
    • 6.3.4 Discussion
  • 6.4 Other Applications
    • 6.4.1 Clocks
    • 6.4.2 Real-Time Scheduling
    • 6.4.3 Multiprocessor and Distributed Scheduling
  • 6.5 Summary
• 7. Measurements and Evaluation
  • 7.1 Measurement Environment
    • 7.1.1 Hardware
    • 7.1.2 Software
  • 7.2 User-Level Measurements
    • 7.2.1 Comparing Synthesis with SUNOS 3.5
    • 7.2.2 Comparing Window Systems
  • 7.3 Detailed Measurements
    • 7.3.1 File and Device I/O
    • 7.3.2 Virtual Memory
    • 7.3.3 Window System
    • 7.3.4 Other Figures
  • 7.4 Experience
    • 7.4.1 Assembly Language
    • 7.4.2 Porting Synthesis to the Sony NEWS Workstation
    • 7.4.3 Architecture Support
  • 7.5 Other Opinions
• 8. Conclusion