1. Operating System Concepts and Techniques Lecture 11 Memory Management-4 M. Naghibzadeh Reference M. Naghibzadeh, Operating System Concepts and Techniques, First ed., iUniverse Inc., 2011. To order: www.iUniverse.com, www.barnesandnoble.com, or www.amazon.comwww.iUniverse.comwww.barnesandnoble.comwww.amazon.com

2. Optimal page replacement policy Although not practical, there is an optimal algorithm Removes the page which will be referred to in the farthest future 2 Page trace23415235421 Page frame 12223232 2121 22323 222i2i 1 Page frame 2334343 3232 33i3i 555i5i 5 Page frame 34646 1717 5353 44242 444i4i 4 Hit?-----++-++- Optimal with a main memory of three frames, H=4, M=7, h=4/11, and m=7/11

3. Multilevel Page-Table So far page table was exempted from page removal With large page tables memory efficiency reduces We have to have a way of knowing which part (page) of page table is in main memory and which is not Two level page table serves this purpose Will discuss details in Windows case study 3

4. Cache memory management Cache is a read/write Random access memory similar to main memory but faster It sits between CPU and main memory It is a temporary place for some main memory locations 4 CPU (registers) Cache memory Main memory Secondary storage Can be write through or non-write through

5. Cache memory management… Most main memory management policies can be adopted to cache However, there are special policies too Information transfer to cache is in block unit, a block is much smaller than a page Let’s assume we have 1Mega bytes of cache, block size of 128 bytes Main memory can be divided into blocks of size 128 bytes An address in main memory can be shown as 5 Row number Column number Offset 12 bits 13 bits 7 bits

6. Cache management With this policy, only one block of each column of main memory is in cache 6 0... 2 13 -1 1 0 1 2. 2 12 -1 2 Are inputs equal? Yes, address in cache No, address not in cache Register array Physical address Main memory Cache memory The structure to check whether a physical address is in cache or not

7. Effective access time Suppose cache success rate is 0.9 Therefore, in only 10% of the times an actual access to main memory will take place. If cache access time is say 20 nanoseconds and main memory access time is 50 nanoseconds, on the average, it will take 0.9*20 + 0.1*(x + 50) nanoseconds to access what we need x is the time that takes to fine out what we need to access is in the cache or not. For x=5 nanosecons effective main memory access time will become 0.9*20 + 0.1*(5 + 50) = 23.5 nanoseconds; much better than 50 nanoseconds 7

8. Windows case study Two-level page table Page size of 4K bytes An address looks like It uses dynamic working set, with default value at the beginning It uses per process page removal Page removal algorithm is FIFO It prefetches pages of main memory 8 31 30 … 22 21 20 … 12 11 10 … 2 1 0 Partial Page table Page number Offset

9. Address translation in Windows 9 Partial page table number Page number Offset Page frame number Offset Logical address Physical address The partial page table corresponding to left most 10 bits of the logical address Present? Yes No Interrupt............ Yes No Interrupt Present?............ Directory

10. Unix case study One-level page table Page size of usually 1K bytes An address looks like It uses dynamic working set, with default value at the beginning It uses global page removal Page removal algorithm is a modified version of clock 10 31 30 … 22 21 20 … 12 11 10 … 2 1 0 Page number Offset

11. Unix page removal algorithm UNIX likes to have a certain minimum number of free page frames all the times, minfree (or low water level) If it becomes lower, page daemon agent will try to free page frames until the number of page frames reaches lotsfree (or high water level) Think of filling up a water tank when the water level is low It uses a modified version of the clock algorithm to decide which pages to remove more than one page is removed every time page daemon agent is activated 11

12. Unix page removal algorithm.. If many frames are scanned by page daemon, but the number of free frames have not reached lotsfree the process swapper complements the page removal process Process swapper will swap out some processes These processes will be swapped back when the situation improves With this view, take one more look at Unix state transition diagram 12

13. Summary Cache memory is part of all computers, MM would be incomplete without talking about cache and the way it is managed It improves effective memory access time depending on cache hit ratio Two case studies were discussed UNIX uses special concepts such as process swapping, lotsfree and minfree Windows, on the other hand, uses two-level page table memory management which allows virtual memory for page tables 13

14. 14 Find out A page replacement policy which always gives the highest page faults The total size of directory and partial page tables for a 4giga bytes program in Windows Effective memory access in Windows MM with two- level page table and cache memory for address traslation In UNIX, what are the rationales for stopping the page daemon agent when “enough” frames are scanned but not enough pages are freed The criteria for selecting a process to swap out in Unix and vice versa

15. 15 Any questions?