Final Topics

• Interacting with I/O devices
  • Device controller, device driver
  • Object-oriented device interfaces (such as block devices)
  • Busy waiting vs. interrupt-driven vs. polling
  • Request buffers, DMA rings
  • Overhead, latency, throughput of I/O interaction methodologies
  • Bottlenecks
• Memory management
  • Internal fragmentation, external fragmentation
  • Address space contents: code, global data, stack, heap
• Virtual memory
  • Page map function
  • Page faults
  • Virtual address space vs. physical address space
  • Architectural support
  • Achieve allocation flexibility
  • Achieve isolation: separate per-process address spaces
  • Achieve illusion of larger memory: swapping
    • Page replacement algorithms: optimal, FIFO, LRU
    • Belady's anomaly
    • Implementing LRU
  • Achieve performance
    • Copy-on-write fork
    • Demand paging
    • Memory mapped files
    • Prefetching
  • Extended page map functions (include access type, access privilege)
  • Architectural implementation: x86
• Disks
  • Platter, surface, track, sector, cylinder
  • Seek, rotational latency, transfer rates
• File systems
  • Kernel memory representation
    • Object-oriented file system interface
    • Inode, file structures
    • Buffer cache
  • Performance: caching, layout, prefetching
  • Robustness (recovery from failures): replication, redundancy, ordering
  • Efficiency: space overhead
  • Data vs. metadata
  • Boot sector, superblock
  • File allocation table
  • Inode design (Unix): free block bitmap, indirect blocks
  • Disk scheduling algorithms: FCFS, Shortest Seek Time First, SCAN/LOOK, Circular LOOK (elevator scheduling)
  • Directory implementation: links, mounting, symbolic links
  • File system correctness/consistency
    • Invariants: No object is used for more than one purpose; Every allocated object is marked allocated; Every free object is marked free; Every referenced object is initialized
    • Fsck
    • Write ordering
    • Journaling: transactions
    • RAID
    • Mean time to failure
• Distributed systems
  • Loosely coupled vs. tightly coupled
  • Distributed shared memory
  • RPC: stubs, automatic stub generators
  • Asynchronous RPC (and AJAX)
  • Writing a server with sockets
  • Server architectures: single process, multiprocess, multithreaded, single process event-driven
  • Receive livelock and interrupts
  • Distributed file systems and NFS
    • NFS and RPC, server design
    • Buffering and caching
    • Consistency models: write-to-read vs. close-to-open
    • Stateless protocol
• Security
  • Prevent unauthorized access & ensure authorized access
  • Defining authorization: principal, resource/object, access right & access matrix
  • Principle of least privilege
  • Unix access matrices (e.g. for processes, for files): user IDs, group IDs
  • Access control lists, capabilities
  • Cryptography: symmetric (private-key), asymmetric (public-key)
  • Replay attack