[UCSD CSE120]内存管理-memory_management

本文是我在上UCSD的 CSE 120: Principles of Operating Systems (Winter 2020) 整理的笔记，这一课主要介绍了操作系统里面对内存的分配以及管理方式

Basic

1. Allocation of memory occurs when

   • new process is created
   • process requests more memory

2. Freeing of memory occurs when

   • process exits
   • process no longer needs memory it requested

3. Memory overview

   • Porcess memory store:

     • Text: code of program
     • Data: static varibales, heap
     • Stack: automatic variables, activation records
     • Other: shared memory regions

   • Process memory address space

     • Address space

       • set of addresses to access memory
       • Typically, linear and sequential
       • 0 to N-1

     • Example (right):

       • Text of size X at 0 ~ X-1
       • Data of size Y at X ~ X+Y-1
       • Stack of size Z at N-Z ~ N-1(grow reversely)

   • Compiler’s model of memory

     • Compiler generates memory address

       • Address ranges of text, data, stack
       • allow data and stack to grow

     • What is not known in compiler

       • Physical memory size
       • Allocated region of physical memory

   • Memory characteristics

     • Size, fixed or variable (max size)
     • Permission: r, w, x

4. Goal: support multiple processes

   • Support programs running “simultaneously”

     • implement process abstraction
     • Multiplex CPU time over all runnable processes
     • Disk r/w speed is low: must keep multiple processes in memory

   • Process requires more than CPU time: memory

5. Memory issues and topics

   • Where should process memories be placed? -> Memory Management
   • How does the compiler model memory? -> Logical memory model, segmentation
   • How to deal with limited physical memory? -> Virtual memory, paging
   • Machanism and Policies

Memory Managemeng Implementation

1. example process

   • Phsical memory starts as one empty “hole”
   • When processes are created, areas get allocated

   • To allocate memory

     • find large enough hole
     • allocate region within hole
     • mostly leaves smaller hole (when the hole size does not exactly match the size of process memory)
     • when process exit (or memeory no longer needed), release that area, which create a new hole, coaleasce with adjacent

   • problem: if there are multiple holes, which to select?

     • First fit: simple, fast

       • consider tradeoff: fit vs. search time
       • memory is cheap, time is expensive

     • Best fit: optimal, must check every hole, leaves very small fragments

     • Worst fit: leaves large fragments, must check every hole

   • complication: is region fixed or variable size?

2. Fragmentation

   • Over time, memory becomes fragmented, there may be signficant unused space (fragmented)

   • Internal fragmentation

     • Unused space within (allocated) block
     • Cannot be allocated to others
     • Can come in handy for growth (for stack or heap)

   • External fragmentation

     • Unused space outside any blocks (holes)
     • Can be allocated (too small/not useful)

   • Approaches

     • Compaction

       • Simple idea
       • Very time consuming

     • Break block (to be allocated) into sub-blocks

       • Easier to fit
       • But complex

     • Use pre-sized holes

       • Same-sized holes:

         • all holes same, easy allocation
         • may be too small which is inflexible

       • Varaiety of sizes (small, medium, large)

         • more flexible
         • complex
         • what should sizes be? how to determine

       • Not adaptable, cause internel fragmentation

3. Some rules

   • 50% Rule: m = n / 2

     • Block: an allocated block
     • Hole: free space between blocks
     • m = number of holes
     • n = number of blocks

   • Unused Memory Rule: f = k / (k + 2)

     • Given average size of blocks and holes are known
     • b = average size of blocks
     • h = average size of holes
     • k = h / b, ratio of average hole-to-block size
     • f = k / (k + 2) is fraction space lost to holes

   • Usage:

     • k = 1, f = 1/3 -> avg hole size = avg block size, 33% waste
     • k = 2, f = 1/2 -> avg hole size = 2 * avg block size, 50% waste
     • k = 8, f = 4/5 -> avg hole size = 8 * avg block size, 80% waste

   • Limits

     • In general, f increases with increasing k

       • as $k \rightarrow \infty, f \rightarrow 1$

     • Alternatively, f decreases with decreasing k

       • as $k \rightarrow 0, f \rightarrow 0$

4. The Buddy System

   • Partition into power-of-2 size chunks

   • Allocation: given request for size r

     find chunk larger than r (else return failure)
     while (r < sizeof(chunk) / 2)
         divide chunk into buddies (each 1/2 size)
     allocate the chunk

   • Free: free the chunk and coalesce with buddy

     free the chunk
     while (buddy is also free)
         coalesce

   • visualization

   • Data structure for buddy-system -> binary tree