530 likes | 729 Views
Principles of Operating Systems: Design & Applications. Chapter 9 Principles of Memory Management. Objectives. understand the relationships among the various types of memory in the memory hierarchy understand the purpose and general idea of address translation
E N D
Principles of Operating Systems: Design & Applications Chapter 9 Principles of Memory Management
Objectives • understand the relationships among the various types of memory in the memory hierarchy • understand the purpose and general idea of address translation • understand segmented address translation • understand paged address translation • be able to identify the application of each of the memory services
Objectives • have a solid mental image of the various approaches to memory layout, both globally and per-process • be able to compare free space representations and apply them at appropriate times • understand the difference between internal and external fragmentation • understand the various memory allocation techniques • understand the various page replacement policies
The Memory Hierarchy • Engineering tradeoff: • Cost and volumeversus • Speed and capacity • For a given point in time, increased speed and capacity require greater cost and volume Principles of Operating Systems: Design & Applications 4
Memory Hierarchy (cont.) Principles of Operating Systems: Design & Applications 5
Address Translation • Relocation • Protection • Convert virtual addresses to physical addresses • Needs to be fast: done in hardware Principles of Operating Systems: Design & Applications 6
Base/Limit Translation Principles of Operating Systems: Design & Applications 7
Segmented Translation Principles of Operating Systems: Design & Applications 8
Page Table Principles of Operating Systems: Design & Applications 9
Two-Level Page Table Principles of Operating Systems: Design & Applications 10
Page Table Entry • Page Frame Number (PFN): becomes upper bits of physical address • Protection bits: determine allowed accesses • Present bit (P): indicates whether page is resident in memory • Dirty bit (D): indicates whether page has been written to • Accessed bit (A): indicates whether page has been referenced Principles of Operating Systems: Design & Applications 11
Translation Lookaside Buffer • Page table lookups are usually memory references • Slow memory references by factor of 2, 3, 4, or even 5 • TLB caches most recent PTEs Principles of Operating Systems: Design & Applications 12
Inverted Page Table • Full page table can be prohibitively large • Often page frames in use are a small fraction of virtual address space • Alternative: translate page frame numbers to page numbers • Use TLB for recent translations Principles of Operating Systems: Design & Applications 13
Memory Services • Explicit allocation: process specifies virtual address range • Implicit allocation: kernel decides virtual range • Garbage collection: identify and free unused blocks • Memory mapped file: file whose contents are directly associated with a region in the virtual address space Principles of Operating Systems: Design & Applications 14
Typical System Memory Layout Principles of Operating Systems: Design & Applications 15
Typical Process Memory Layout Principles of Operating Systems: Design & Applications 16
Free Space Management • Free bit map: • Good for fixed-sized blocks, like pages • Low overhead • Free list: • Sometimes tree or hash table • Standard trick: store overhead structures in free blocks themselves Principles of Operating Systems: Design & Applications 17
Free List Example Principles of Operating Systems: Design & Applications 18
Fragmentation • External fragmentation: wasted space outside allocation units • Internal fragmentation: wasted space within allocation units Principles of Operating Systems: Design & Applications 19
Partitions • Statically sized • Allocated in whole • Usually set up at boot time • Usually administratively determined Principles of Operating Systems: Design & Applications 20
Variable-Sized Allocation • Among free blocks at least as large as the request: • First fit: pick the first • Next fit: pick the next • Best fit: pick the smallest • Worst fit: pick the biggest Principles of Operating Systems: Design & Applications 21
First Fit Example Principles of Operating Systems: Design & Applications 22
Next Fit Example Principles of Operating Systems: Design & Applications 23
Best Fit Example Principles of Operating Systems: Design & Applications 24
Worst Fit Example Principles of Operating Systems: Design & Applications 25
Buddy System • Allocation algorithm: • If n is less than the smallest allocation unit, set n to that smallest size • Round n up to the nearest power of 2 times the smallest allocation unit • If there is no block of the needed size, recursively allocate the next larger size and split it into two • Return the first block of the required size Principles of Operating Systems: Design & Applications 26
Buddy System (cont.) Principles of Operating Systems: Design & Applications 27
Overallocation • Swapping: copying full processes between main memory and backing store • Segment swapping: copying individual segments between main memory and backing store • Paging (page swapping): copying individual pages between main memory and backing store Principles of Operating Systems: Design & Applications 28
Paging Concepts • Demand paging: reading pages into memory in response to page faults • Page fault: an interrupt that occurs when a translation is not valid, often because the page is not resident in memory • Prepaging: loading pages in anticipation of later needing them Principles of Operating Systems: Design & Applications 29
Paging Concepts (cont.) • Page reference string: list of page numbers in the order in which they are referenced • Global page replacement policy: page replacement policy that selects pages without regard to the processes that own them • Local page replacement policy: page replacement policy that selects pages only from a given process Principles of Operating Systems: Design & Applications 30
Belady's Min • Optimal page replacement policy • Pick the page for which it will be longest before it's used again: • Unrealizable—noncausal • Standard of comparison for realizable policies Principles of Operating Systems: Design & Applications 31
First In, First Out • Simple page replacement policy • Select the page that has been resident for the longest period of time • Implemented with a queue • Can “jump the gun” and swap out frequently used pages just because they've been there a long time Principles of Operating Systems: Design & Applications 32
Second Chance • Address FIFO weakness • If the process at the head of the queue has A=1, clear A and reinsert it at the tail, otherwise, swap it out • Allows frequently used pages to stay in memory for a long time • If all pages have A=1, then all get cleared and the oldest is swapped out—reduces to FIFO Principles of Operating Systems: Design & Applications 33
Clock Algorithm • Implementation of second chance • Pages are arranged in a circle and a “clock hand” sweeps them • At each step, if A=1, clear A and advance, otherwise, copy the page out Principles of Operating Systems: Design & Applications 34
Two Hand Clock • If there are many pages, nearly all may have A=1 • Reduce time between clearing A and testing • Two “clock hands” move in lockstep: first clears, second tests Principles of Operating Systems: Design & Applications 35
Not Recently Used • If D=0 (M=0), the page need not be written; the new one can be copied directly in to the frame • Pages with D=0, can be swapped faster • Let AD (AM) be a two bit number: 00, 01, 10, 11 (0, 1, 2, 3) • Select a page from the set with the lowest number Principles of Operating Systems: Design & Applications 36
Least Recently Used • Pages not used for a long time may not be needed for a long time • Least recently used is a realizable approximation to Belady's min • Rarely implemented directly—hardware rarely records access time • Approximate by periodically clearing P and recording the time of the page fault before resetting P Principles of Operating Systems: Design & Applications 37
Not Frequently Used • Approximate LRU by looking at recent frequency of usage • For each sweep of pages compute:where • Pick a page with the lowest value of c Principles of Operating Systems: Design & Applications 38
Paging Policies Comparison Principles of Operating Systems: Design & Applications 39
Paging Policies Comparison (cont.) Principles of Operating Systems: Design & Applications 40
The Working Set • Set of pages currently “in use” by a process • Keeping more pages is unneeded • Keeping fewer pages leads to a large number of page faults • Attempt to keep the number of pages between two thresholds • Page locally if above the upper • Swap out whole processes that are below the lower Principles of Operating Systems: Design & Applications 41
Page Fault Frequency • Technique to adjust working set thresholds • Keep an average of the process's page fault frequency • If too high, increase the working set size • If very low, decrease the working set size Principles of Operating Systems: Design & Applications 42
Copy on Write (COW) • Technique to avoid copying pages unnecessarily • Map both process's pages to same frames • Make the “copied” pages read-only • When a process writes, the fault triggers: • Make a copy of the page • Map the two process's pages to separate copies • Change the access back to its original value Principles of Operating Systems: Design & Applications 43
Average Access Time • Average access time is given by:where is the time to access memory, is the time to access the disk, and is the fraction of accesses causing page faults • must be very small to keep average time reasonable Principles of Operating Systems: Design & Applications 44
Thrashing • Effect of active working sets being larger than the physical memory • The system spends more time swapping than computing • System performance intolerably slow • Can be partially ameliorated with two-level scheduling Principles of Operating Systems: Design & Applications 45
Belady's Anomaly • Phenomenon where adding more page frames leads to more page faults • Some policies, such as FIFO, are susceptible • Some policies, such as LRU, are immune Principles of Operating Systems: Design & Applications 46
Belady's Anomaly Principles of Operating Systems: Design & Applications 47
Summary • Memory hierarchy gives perspective on tradeoffs of speed and size • Address translation is needed for relocation and protection • Translation can be in terms of segments, using base and limit registers • Translation can be in term of pages using page tables containing page table entries Principles of Operating Systems: Design & Applications 48
Summary (cont.) • TLB caches recently used PTEs • Both system and per-process memory layouts start at address 0 • Typically process data segments grow up toward stack and stack down toward data • Free memory can be represented by bitmaps and by lists • Free lists typically keep pointers and sizes inside the free blocks Principles of Operating Systems: Design & Applications 49
Summary (cont.) • Fragmentation reduces usable memory • Partitioning is a simple and old technique being used again with virtualization • Variable-sized blocks can be allocated by first fit, next fit, best fit, worst fit, or buddy system • Memory can be overallocated by using swapping, segment swapping, or page swapping Principles of Operating Systems: Design & Applications 50