1. 1 Virtual Memory

2. Cache memory: provides illusion of very high speed Virtual memory: provides illusion of very large size Main memory: reasonable cost, but slow & small Memory Hierarchy Summary

3. Virtual Memory  Program addresses only logical addresses  Hardware maps logical addresses to physical addresses  Only part of a process is loaded into memory  process may be larger than main memory  additional processes allowed in main memory since only part of each process needs to be in physical memory  memory loaded/unloaded as the programs execute  Real Memory – The physical memory occupied by a program (frames)  Virtual memory – The larger memory space perceived by the program (pages) Virtual Memory

4. Virtual Memory That is Larger Than Physical Memory

5. Virtual Memory  Principle of Locality – A program tends to reference the same items - even if same item not used, nearby items will often be referenced  Resident Set – Those parts of the program being actively used (remaining parts of program on disk)  Thrashing – Constantly needing to get pages off secondary storage  happens if the O.S. throws out a piece of memory that is about to be used  can happen if the program scans a long array – continuously referencing pages not used recently  O.S. must watch out for this situation! Virtual Memory

6.  Decisions about virtual memory:  Fetch Policy – when to bring a page in?  When needed or in anticipation of need?  Placement – where to put it?  Replacement – what to unload to make room for a new page?  Resident Set Management – how many pages to keep in memory?  Fixed # of pages or variable?  Reassign pages to other processes?  Cleaning Policy – when to write a page to disk?  Load Control – degree of multiprogramming?

7. Paging and Virtual Memory  Large logical memory – small real memory  Demand paging allows  size of logical address space not constrained by physical memory  higher utilization of the system  Paging implementation  frame allocation  how many per process?  page replacement  how do we choose a frame to replace?

8. Demand Paging  Bring a page into memory only when it is needed.  Less I/O needed  Less memory needed  Faster response  More users  Page is needed  reference to it  invalid reference  abort  not-in-memory  bring to memory

9. Transfer of a Paged Memory to Contiguous Disk Space

10. Page Table of Demand Paging

11. Page Fault  If there is ever a reference to a page, first reference will trap to OS  page fault  OS looks at another table to decide:  Invalid reference  abort.  Just not in memory.  Get empty frame.  Swap page into frame.  Reset tables, validation bit = 1.  Restart instruction: Least Recently Used  block move  auto increment/decrement location

12. Steps in Handling a Page Fault

13. What happens if there is no free frame?  Page replacement – find some page in memory, but not really in use, swap it out.  algorithm  performance – want an algorithm which will result in minimum number of page faults.  Same page may be brought into memory several times.

14. Demand Paging  Paged memory combined with swapping  Processes reside in main/secondary memory  Could also be termed as lazy swapping  bring pages into memory only when accessed  What about at context switch time?  could swap out entire process  restore page state as remembered  anticipate which pages are needed

15. Page Replacement  Demand paging allows us to over-allocate  no free frames – must implement frame replacement  Frame replacement  select a frame (victim)  write the victim frame to disk  read in new frame  update page tables  restart process

16. Page Replacement  If no frames are free, two page transfers  doubles page fault service time  Reduce overhead using dirty bit  dirty bit is set whenever a page is modified  if dirty, write page, else just throw it out

17. Paging Implementation (continued … )  Must be able to restart a process at any time  instruction fetch  operand fetch  operand store (any memory reference)  Consider simple instruction  Add C,A,B (C = A + B)  All operands on different pages  Instruction not in memory  4 possible page faults )-: slooooow :-(

18. Performance of Demand Paging  Page Fault Rate 0  p  1.0  if p = 0 no page faults  if p = 1, every reference is a fault  Effective Access Time (EAT) EAT = (1 – p) x memory access + p (page fault overhead + [swap page out ] + swap page in + restart overhead)

19. Demand Paging Example  Memory access time = 1 microsecond  50% of the time the page that is being replaced has been modified and therefore needs to be swapped out.  Swap Page Time = 10 msec = 10,000 microsecond EAT = (1 – p) x 1 + p (15000) =1 + 15000p (in microsecond)

20. Performance Example  Paging Time …  Disk latency8 milliseconds  Disk seek15 milliseconds  Disk transfer time1 millisecond  Total paging time~25 milliseconds  Could be longer due to  device queueing time  other paging overhead

21. Paging Performance (continued … )  Effective access time: EAT = (1 - p)  ma + p  pft where: p is probability of page fault ma is memory access time pft is page fault time

22. Paging Performance (continued … )  Effective access time with 100 ns memory access and 25 ms page fault time: EAT = (1 - p)  ma + p  pft = (1 - p)  100 + p  25,000,000 = 100 + 24,999,900  p  What is the EAT if p = 0.001 (1 out of 1000)?  100 + 24999,990  0.001 = 25 microseconds  250 times slowdown!  How do we get less than 10% slowdown?  100 + 24999,990  p  1.10  100 ns = 110 ns  Less than 1 out of 2,500,000 accesses fault

23. Paging Improvements  Paging needs to be as fast as possible  Disk access time is faster if:  use larger blocks  no file table lookup or other indirect lookup  binary boundaries  Most systems have a separate swap space  Copy entire file image into swap at load time  Demand page  Or …  Demand pages initially from the file system  Write pages to swap as they are needed

24. Fetch Policy  Demand paging means that a process starts slowly.  produces a flurry of page faults early, then settles down  Locality means a smaller number of pages per process are needed.  desired set of pages should be in memory - working set  Prepaging means bringing in pages that are likely to be used in the near future.  try to take advantage of disk characteristics  generally more efficient to load several consecutive sectors/pages than individual sectors due to seek, rotational latency  hard to correctly guess which pages will be referenced  easier to guess at program startup  may load unnecessary pages

25. Placement Policies  Where to put the page  trivial in a paging system – can be placed anywhere  Best-fit, First-Fit, or Next-Fit can be used with segmentation  is a concern with distributed systems

26. Replacement Policies  Replacement Policy  which page to replace when a new page needs to be loaded  tends to combine several things:  how many page frames are allocated  replace only a page in the current process or from all processes? (Resident Set Management)  from pages being considered, selecting one page to be replaced  Frame Locking  require a page to stay in memory  O.S. Kernel and Interrupt Handlers  real-Time processes  other key data structures  implemented by bit in data structures