Assignment 10, 11 Synchronization; atomic operations and linked list with synchronization, vm.dirty_ratio Test

[jimin-kiim]

Assignment 10 - Synchronization; atomic operations and linked list with synchronization

Atomic operations

• performing atomic operations using atomic instructions in a simple kernel module with four kernel threads.
  • fetch-and-add
  • test-and-set
  • compare-and-swap
• Each thread should increase the shared resource counter by 1 and print it.

Linked list with synchronization

• Four kernel thread runs simultaneously.
• Each kernel thread performs Insert, Search, Delete operations to the global linked list, which is shared by all threads.
• To protect global linked list from corruption by concurrent access of the threads, single global lock is used.
  • spinlock: the most common lock used in linux kernel
    • simply waits in a loop (spin) while repeatedly checking if the lock is available
    • it doesn't need any context switching
    • busy waiting: when the thread remains active but is not performing a useful task. it can lead to the waste of CPU cycles
    • it's usually used when each thread expects to own a lock for a short time
  • mutex (mutual exclusion): a binary semaphore
    • sleeping lock.
    • mutex / semaphore is usually used when each thread expects to own a lock for a long time or it needs to sleep while holding lock
    • when mutex / semaphore is used, the interrupt is blocked.
  • rw semaphore: a multiple readers and single writer lock mechanism.
    • semaphore cannot be used while holding a spin lock.
• Each thread performs Insert, Search, Delete for its own data range bound.

Assignment 11 - vm.dirty_ratio Test

• run Fio test with
  • vm.dirty_ratio = 20 (default)
  • vm.dirty_ratio = 40
  • vm.dirty_ratio = 100
• dirty ratio: the percentage when the generator starts writeback of dirty pages
• fio: flexible i/o test. tool for measuring performance