System Performance Tuning. 2nd Edition Gian-Paolo D. Musumeci, Mike Loukides

Spis treści książki

• System Performance Tuning, 2nd Edition
  • SPECIAL OFFER: Upgrade this ebook with OReilly
  • Preface
    • Who Should Buy This Book?
    • A Note on Coverage
    • How to Read This Book
      • This Book as a Story
      • This Book as a Reference
    • Organization
    • Typographic Conventions
    • Comments and Questions
    • Personal Comments and Acknowledgments
      • Acknowledgments from Mike Loukides
  • 1. An Introduction to Performance Tuning
    • 1.1. An Introduction to Computer Architecture
      • 1.1.1. Levels of Transformation
        • 1.1.1.1. Software: algorithms and languages
        • 1.1.1.2. The Instruction Set Architecture
        • 1.1.1.3. Hardware: microarchitecture, circuits, and devices
      • 1.1.2. The von Neumann Model
      • 1.1.3. Caches and the Memory Hierarchy
      • 1.1.4. The Benefits of a 64-Bit Architecture
        • 1.1.4.1. What does it mean to be 64-bit?
        • 1.1.4.2. Performance ramifications
    • 1.2. Principles of Performance Tuning
      • 1.2.1. Principle 0: Understand Your Environment
      • 1.2.2. Principle 1: TANSTAAFL!
      • 1.2.3. Principle 2: Throughput Versus Latency
      • 1.2.4. Principle 3: Do Not Overutilize a Resource
      • 1.2.5. Principle 4: Design Tests Carefully
    • 1.3. Static Performance Tuning
      • 1.3.1. Other Miscellaneous Things to Check
    • 1.4. Concluding Thoughts
  • 2. Workflow Management
    • 2.1. Workflow Characterization
      • 2.1.1. Simple Commands
      • 2.1.2. Process Accounting
        • 2.1.2.1. Enabling process accounting
        • 2.1.2.2. Reviewing accounting records
      • 2.1.3. Automating sar
        • 2.1.3.1. Enabling sar
        • 2.1.3.2. Retrieving data
      • 2.1.4. Virtual Adrian
      • 2.1.5. Network Pattern Analysis
        • 2.1.5.1. Pattern 1: request-response
        • 2.1.5.2. Pattern 1B: inverse request-response
        • 2.1.5.3. Pattern 2: data transfer
        • 2.1.5.4. Pattern 3: message passing
        • 2.1.5.5. Packet size distributions
    • 2.2. Workload Control
      • 2.2.1. Education
        • 2.2.1.1. Usage and performance agreements
      • 2.2.2. The maxusers and the pt_cnt Parameters
      • 2.2.3. Limiting Users
        • 2.2.3.1. Quotas
        • 2.2.3.2. Environmental limits
      • 2.2.4. Complex Environments
    • 2.3. Benchmarking
      • 2.3.1. MIPS and Megaflops
        • 2.3.1.1. MIPS
        • 2.3.1.2. Megaflops
      • 2.3.2. Component-Specific Benchmarks
        • 2.3.2.1. Linpack
        • 2.3.2.2. SPECint and SPECfp
      • 2.3.3. Commercial Workload Benchmarks
        • 2.3.3.1. TPC
        • 2.3.3.2. SPECweb99
      • 2.3.4. User Benchmarks
        • 2.3.4.1. Choose your problem set
        • 2.3.4.2. Choose your runtime
        • 2.3.4.3. Automate heavily
        • 2.3.4.4. Set benchmark runtime rules
    • 2.4. Concluding Thoughts
  • 3. Processors
    • 3.1. Microprocessor Architecture
      • 3.1.1. Clock Rates
      • 3.1.2. Pipelining
        • 3.1.2.1. Variable-length instructions
        • 3.1.2.2. Branches
      • 3.1.3. The Second Generation of RISC Processor Design
    • 3.2. Caching
      • 3.2.1. The Cache Hierarchy
      • 3.2.2. Cache Organization and Operation
      • 3.2.3. Associativity
      • 3.2.4. Locality and "Cache-Busters"
        • 3.2.4.1. Unit stride
        • 3.2.4.2. Linked lists
        • 3.2.4.3. Cache-aligned block copy problems
      • 3.2.5. The Cache Size Anomaly
    • 3.3. Process Scheduling
      • 3.3.1. The System V Model: The Linux Model
        • 3.3.1.1. Finding a processs priority
        • 3.3.1.2. Adjusting a process's effective priority
        • 3.3.1.3. Modifications for SMP systems
      • 3.3.2. Multilayered Scheduling Classes: The Solaris Model
        • 3.3.2.1. The Solaris threading model
        • 3.3.2.2. Scheduling classes
        • 3.3.2.3. The dispatcher
        • 3.3.2.4. Checking a process's priority
        • 3.3.2.5. Tuning the dispatch tables
        • 3.3.2.6. Adjusting priorities
    • 3.4. Multiprocessing
      • 3.4.1. Processor Communication
        • 3.4.1.1. Buses
        • 3.4.1.2. Crossbars
        • 3.4.1.3. UltraSPARC-III systems: Fireplane
        • 3.4.1.4. "Interconnectionless" architectures
      • 3.4.2. Operating System Multiprocessing
      • 3.4.3. Threads
      • 3.4.4. Locking
      • 3.4.5. Cache Influences on Multiprocessor Performance
    • 3.5. Peripheral Interconnects
      • 3.5.1. SBus
        • 3.5.1.1. Clock speed
        • 3.5.1.2. Burst transfer size
        • 3.5.1.3. Transfer mode
        • 3.5.1.4. Summary of SBus implementations
        • 3.5.1.5. SBus card utilization
      • 3.5.2. PCI
        • 3.5.2.1. PCI bus transactions
        • 3.5.2.2. CompactPCI
      • 3.5.3. A Summary of Peripheral Interconnects
      • 3.5.4. Interrupts in Linux
      • 3.5.5. Interrupts in Solaris
    • 3.6. Processor Performance Tools
      • 3.6.1. The Load Average
      • 3.6.2. Process Queues
      • 3.6.3. Specific Breakdowns
      • 3.6.4. Multiprocessor Systems
      • 3.6.5. top and prstat
      • 3.6.6. Lock Statistics
      • 3.6.7. Controlling Processors in Solaris
        • 3.6.7.1. psrinfo
        • 3.6.7.2. psradm
        • 3.6.7.3. psrset
      • 3.6.8. Peripheral Interconnect Performance Tools
      • 3.6.9. Advanced Processor Performance Statistics
    • 3.7. Concluding Thoughts
  • 4. Memory
    • 4.1. Implementations of Physical Memory
    • 4.2. Virtual Memory Architecture
      • 4.2.1. Pages
      • 4.2.2. Segments
      • 4.2.3. Estimating Memory Requirements
      • 4.2.4. Address Space Layout
      • 4.2.5. The Free List
        • 4.2.5.1. Virtual memory management in Linux
      • 4.2.6. Page Coloring
      • 4.2.7. Transaction Lookaside Buffers (TLB)
    • 4.3. Paging and Swapping
      • 4.3.1. The Decline and Fall of Interactive Performance
      • 4.3.2. Swap Space
        • 4.3.2.1. Anonymous memory
        • 4.3.2.2. Sizing swap space
        • 4.3.2.3. Organizing swap space
        • 4.3.2.4. Swapfiles
    • 4.4. Consumers of Memory
      • 4.4.1. Filesystem Caching
      • 4.4.2. Filesystem Cache Writes: fsflush and bdflush
        • 4.4.2.1. Solaris: fsflush
        • 4.4.2.2. Linux: bdflush
      • 4.4.3. Interactions Between the Filesystem Cache and Memory
        • 4.4.3.1. Priority paging
        • 4.4.3.2. Cyclic caching
      • 4.4.4. Interactions Between the Filesystem Cache and Disk
    • 4.5. Tools for Memory Performance Analysis
      • 4.5.1. Memory Benchmarking
        • 4.5.1.1. STREAM
        • 4.5.1.2. lmbench
      • 4.5.2. Examining Memory Usage System-Wide
        • 4.5.2.1. vmstat
        • 4.5.2.2. sar
        • 4.5.2.3. memstat
      • 4.5.3. Examining Memory Usage of Processes
        • 4.5.3.1. Solaris tools
      • 4.5.4. Linux Tools
    • 4.6. Concluding Thoughts
  • 5. Disks
    • 5.1. Disk Architecture
      • 5.1.1. Zoned Bit Rate (ZBR) recording
        • 5.1.1.1. Disk caches
      • 5.1.2. Access Patterns
      • 5.1.3. Reads
      • 5.1.4. Writes
        • 5.1.4.1. UFS write throttling
      • 5.1.5. Performance Specifications
        • 5.1.5.1. One million bytes is a megabyte?
        • 5.1.5.2. Burst speed versus internal transfer speed
        • 5.1.5.3. Internal transfer speed versus actual speed
        • 5.1.5.4. Average seek time
        • 5.1.5.5. Storage capacity and access capacity
    • 5.2. Interfaces
      • 5.2.1. IDE
        • 5.2.1.1. Improving IDE performance in Linux
        • 5.2.1.2. Limitations of IDE drives
      • 5.2.2. IPI
      • 5.2.3. SCSI
        • 5.2.3.1. Multi-initiator SCSI
        • 5.2.3.2. Bus transactions
        • 5.2.3.3. Synchronous versus asynchronous transfers
        • 5.2.3.4. Termination
        • 5.2.3.5. Command queuing
        • 5.2.3.6. Differential signaling
        • 5.2.3.7. Bus utilization
        • 5.2.3.8. Mixing different speed SCSI devices
        • 5.2.3.9. SCSI implementations
      • 5.2.4. Fibre Channel
      • 5.2.5. IEEE 1394 (FireWire)
      • 5.2.6. Universal Serial Bus (USB)
    • 5.3. Common Performance Problems
      • 5.3.1. High I/O Skew
      • 5.3.2. Memory-Disk Interactions
      • 5.3.3. High Service Times
    • 5.4. Filesystems
      • 5.4.1. vnodes, inodes, and rnodes
        • 5.4.1.1. The directory name lookup cache (DNLC)
      • 5.4.2. The Unix Filesystem (UFS)
        • 5.4.2.1. inode density
        • 5.4.2.2. Filesystem cluster size
        • 5.4.2.3. Minimum free space
        • 5.4.2.4. Rotational delay
        • 5.4.2.5. fstyp and tunefs
        • 5.4.2.6. Bypassing memory caching
        • 5.4.2.7. The inode cache
        • 5.4.2.8. The buffer cache
      • 5.4.3. Logging Filesystems
        • 5.4.3.1. Solstice:DiskSuite
        • 5.4.3.2. Solaris
      • 5.4.4. The Second Extended Filesystem (EXT2)
      • 5.4.5. The Third Extended Filesystem (EXT3)
        • 5.4.5.1. Tuning the elevator algorithm
        • 5.4.5.2. Choosing a journaling mode
        • 5.4.5.3. Transitioning from ext2 to ext3
      • 5.4.6. The Reiser Filesystem (ReiserFS)
        • 5.4.6.1. Tail packing
      • 5.4.7. The Journaled Filesystem (JFS)
      • 5.4.8. The Temporary Filesystem (tmpfs)
      • 5.4.9. Veritas VxFS
      • 5.4.10. Caching Filesystems (CacheFS)
        • 5.4.10.1. Minimizing seek times by filesystem layout
    • 5.5. Tools for Analysis
      • 5.5.1. Enabling Disk Caches
      • 5.5.2. Disk Performance Benchmarking
        • 5.5.2.1. hdparm
        • 5.5.2.2. tiobench
        • 5.5.2.3. iozone
      • 5.5.3. Second-Time-Through Improvements?
      • 5.5.4. Using iostat
        • 5.5.4.1. Historical limitations: iostat and queuing terminology
      • 5.5.5. Using sar
      • 5.5.6. I/O Tracing
        • 5.5.6.1. Using the kernel probes
        • 5.5.6.2. Using process filtering
        • 5.5.6.3. Restarting prex
    • 5.6. Concluding Thoughts
  • 6. Disk Arrays
    • 6.1. Terminology
    • 6.2. RAID Levels
      • 6.2.1. RAID 0: Striping
      • 6.2.2. RAID 1: Mirroring
      • 6.2.3. RAID 2: Hamming Code Arrays
      • 6.2.4. RAID 3: Parity-Protected Striping
      • 6.2.5. RAID 4: Parity-Protected Striping with Independent Disks
      • 6.2.6. RAID 5: Distributed, Parity-Protected Striping
      • 6.2.7. RAID 10: Mirrored Striping
    • 6.3. Software Versus Hardware
      • 6.3.1. Software
      • 6.3.2. Hardware
        • 6.3.2.1. RAID overlap
    • 6.4. A Summary of Disk Array Design
      • 6.4.1. Choosing a RAID Level
    • 6.5. Software RAID Implementations
      • 6.5.1. Solaris: Solstice DiskSuite
        • 6.5.1.1. State databases
        • 6.5.1.2. RAID 0: stripes
        • 6.5.1.3. RAID 1: mirrors
        • 6.5.1.4. RAID 5 arrays
        • 6.5.1.5. Hot spare pools
      • 6.5.2. Linux: md
        • 6.5.2.1. Persistent superblocks
        • 6.5.2.2. Chunk size
        • 6.5.2.3. Linear mode
        • 6.5.2.4. RAID 0: stripes
        • 6.5.2.5. RAID 1: mirrors
        • 6.5.2.6. RAID 5 arrays
        • 6.5.2.7. Creating the array
        • 6.5.2.8. Autodetection
        • 6.5.2.9. Booting from an array device
    • 6.6. RAID Recipes
      • 6.6.1. Attribute-Intensive Home Directories
      • 6.6.2. Data-Intensive Home Directories
      • 6.6.3. High Performance Computing
      • 6.6.4. Databases
      • 6.6.5. Case Study: Applications Doing Large I/O
    • 6.7. Concluding Thoughts
  • 7. Networks
    • 7.1. Network Principles
      • 7.1.1. The OSI Model
    • 7.2. Physical Media
      • 7.2.1. UTP
        • 7.2.1.1. A note on terminology: plugs and jacks
      • 7.2.2. Fiber
    • 7.3. Network Interfaces
      • 7.3.1. Ethernet
        • 7.3.1.1. Fundamentals of Ethernet signaling
        • 7.3.1.2. Topologies
        • 7.3.1.3. 10BASE-T
        • 7.3.1.4. 100BASE-T4
        • 7.3.1.5. 100BASE-TX
        • 7.3.1.6. Gigabit Ethernet topologies
        • 7.3.1.7. The 5-4-3 rule
        • 7.3.1.8. Collisions
        • 7.3.1.9. Autonegotiation
        • 7.3.1.10. Displaying and setting modes
      • 7.3.2. FDDI
      • 7.3.3. ATM
      • 7.3.4. Ethernet Versus ATM/FDDI
    • 7.4. Network Protocols
      • 7.4.1. IP
        • 7.4.1.1. Fragmentation
        • 7.4.1.2. Time-to-live
        • 7.4.1.3. Protocols
        • 7.4.1.4. IP addresses
        • 7.4.1.5. Classful addressing
        • 7.4.1.6. Subnetting classful networks
        • 7.4.1.7. Moving to a classless world
        • 7.4.1.8. Routing
      • 7.4.2. TCP
        • 7.4.2.1. Connection initiation and SYN flooding
        • 7.4.2.2. Path MTU discovery and the maximum segment size
        • 7.4.2.3. Buffers, watermarks, and windows
        • 7.4.2.4. Retransmissions
        • 7.4.2.5. Deferring acknowledgments
        • 7.4.2.6. Window congestion and the slow start algorithm
        • 7.4.2.7. TCP timers and intervals
        • 7.4.2.8. The Nagle algorithm
      • 7.4.3. UDP
      • 7.4.4. TCP Versus UDP for Network Transport
    • 7.5. NFS
      • 7.5.1. Characterizing NFS Activity
      • 7.5.2. Tuning Clients
        • 7.5.2.1. Obtaining statistics for an NFS-mounted filesystem
        • 7.5.2.2. The rnode cache
        • 7.5.2.3. Tuning NFS clients for bursty transfers
        • 7.5.2.4. Tuning NFS clients for sequential transfer
      • 7.5.3. Tuning Servers
        • 7.5.3.1. Designing disk subsystems for NFS servers
        • 7.5.3.2. NVRAM caching
        • 7.5.3.3. Memory requirements
        • 7.5.3.4. The two basic types of NFS servers
        • 7.5.3.5. Tuning the number of NFS threads
        • 7.5.3.6. Adjusting the buffer cache
        • 7.5.3.7. The maxusers parameter
        • 7.5.3.8. The directory name lookup cache (DNLC)
        • 7.5.3.9. The inode cache
        • 7.5.3.10. Observing NFS server performance with nfsstat
      • 7.5.4. Wide Area Networks and NFS
    • 7.6. CIFS via Unix
    • 7.7. Concluding Thoughts
  • 8. Code Tuning
    • 8.1. The Two Critical Approaches
      • 8.1.1. String Searching Algorithms
        • 8.1.1.1. Algorithm 1: naive searching
        • 8.1.1.2. Algorithm 2: Knuth-Morris-Pratt searching
        • 8.1.1.3. Algorithm 3: Boyer-Moore searching
      • 8.1.2. Caveats in Optimization
    • 8.2. Techniques for Code Analysis
      • 8.2.1. Application Timing: time, timex, and ptime
        • 8.2.1.1. time
        • 8.2.1.2. timex
        • 8.2.1.3. ptime
        • 8.2.1.4. Mechanisms of timing
      • 8.2.2. Timing-Specific Code Sections
        • 8.2.2.1. Timing via gethrtime
        • 8.2.2.2. Timing via the TICK register
      • 8.2.3. Probe-Based Analysis: Solaris TNF
        • 8.2.3.1. Inserting probes
        • 8.2.3.2. Caveats
      • 8.2.4. Profiler-Based Analysis: gprof
        • 8.2.4.1. Implementing profiling
        • 8.2.4.2. Compiling with profiling
        • 8.2.4.3. Execution with profiling
        • 8.2.4.4. Profile analysis
        • 8.2.4.5. Caveats
    • 8.3. Optimization Patterns
      • 8.3.1. Arithmetic
      • 8.3.2. Loops
      • 8.3.3. Strings
    • 8.4. Interacting with Compilers
      • 8.4.1. Typical Optimizations: -fast
      • 8.4.2. Optimization Level: -xO
      • 8.4.3. Specifying Instruction Set Architecture: -xarch
      • 8.4.4. Specifying Processor Architecture: -xchip
      • 8.4.5. Function Inlining: -xinlining and -xcrossfile
      • 8.4.6. Data Dependency Analysis: -xdepend
      • 8.4.7. Vector Operations: -xvector
      • 8.4.8. Default Floating Point Constant Size: -xsfpconst
      • 8.4.9. Data Prefetching: -xprefetch
      • 8.4.10. Quick and Dirty Compiler Flags
      • 8.4.11. Profiling Feedback
    • 8.5. Concluding Thoughts
  • 9. Instant Tuning
    • 9.1. Top Five Tuning Tips
      • 9.1.1. Where Is the Disk Bottleneck?
      • 9.1.2. Do You Have Enough Memory?
      • 9.1.3. Are the Processors Overloaded?
      • 9.1.4. Are Processes Blocked on Disk I/O?
      • 9.1.5. Does System Time Heavily Dominate User Time?
    • 9.2. Instant Tuning Recipes
      • 9.2.1. Single-User Development Workstations
        • 9.2.1.1. Filesystems
        • 9.2.1.2. Swap space
        • 9.2.1.3. Kernel tuning
      • 9.2.2. Workgroup Servers
        • 9.2.2.1. Memory
        • 9.2.2.2. Disks
        • 9.2.2.3. Filesystems
        • 9.2.2.4. Swap space
        • 9.2.2.5. Optimizing NFS
        • 9.2.2.6. Kernel tuning
      • 9.2.3. Web Servers
        • 9.2.3.1. Memory
        • 9.2.3.2. Disks
        • 9.2.3.3. Filesystems
        • 9.2.3.4. Swap space
        • 9.2.3.5. Networks
        • 9.2.3.6. Kernel tuning
        • 9.2.3.7. Special case: proxy servers
  • About the Authors
  • Colophon
  • SPECIAL OFFER: Upgrade this ebook with OReilly

pokaż cały spis treści

ukryj recenzje