CS 153 Design of Operating Systems Spring 18

Transcription

1 CS 53 Design of Operating Systems Spring 8 Lectre 2: Virtal Memory Instrctor: Chengy Song Slide contribtions from Nael Ab-Ghazaleh, Harsha Madhyvasta and Zhiyn Qian

2 Recap: cache Well-written programs exhibit a property called locality, this is the fondation of caching Basic concepts of cache Cache hit Cache miss Cache replacement policy is important CS 53 Lectre 2 Virtal Memory 2

3 Today Virtal memory Memory as cache Page/cache replacement policy CS 53 Lectre 2 Virtal Memory 3

4 VM as a Tool for Caching Virtal memory is an array of N contigos bytes stored on disk. The contents of the array on disk are cached in physical memory (DRAM cache) These cache blocks are called pages (size is P = 2 p bytes) Virtal memory Physical memory VP VP Unallocated Cached Uncached Unallocated Empty Empty PP PP Cached Uncached Empty n-p - Cached Uncached N- M- PP 2 m-p - Virtal pages (VPs) stored on disk Physical pages (PPs) cached in DRAM 4

5 Page Table Setp Valid PTEs map virtal pages to physical pages. Invalid PTEs map virtal pages to disk blocks PTE Valid Physical page nmber or disk address nll Physical memory VP VP 7 VP 4 PP PP 3 nll Virtal memory (disk) PTE 7 VP Memory resident page table VP 3 VP 4 CS 53 Lectre 2 Virtal Memory VP 6 5 VP 7

6 Page/Cache Hit Page hit: reference to VM word that is in physical memory (DRAM cache hit) Virtal address PTE PTE 7 Valid Physical page nmber or disk address nll nll Memory resident page table Physical memory VP VP 7 VP 4 Virtal memory (disk) VP VP 3 VP 4 PP PP 3 VP 6 CS 53 Lectre 2 Virtal Memory VP 7 6

7 Page Falt (Cache miss) Page falt: reference to VM word that is not in physical memory (DRAM cache miss) Virtal address PTE PTE 7 Valid Physical page nmber or disk address nll nll Memory resident page table Physical memory VP VP 7 VP 4 Virtal memory (disk) VP VP 3 VP 4 PP PP 3 VP 6 CS 53 Lectre 2 Virtal Memory VP 7 7

8 Handling Page Falt () Page miss cases page falt (an exception) Virtal address PTE Valid Physical page nmber or disk address nll Physical memory VP VP 7 VP 4 PP PP 3 nll Virtal memory (disk) PTE 7 Memory resident page table VP VP 3 VP 4 VP 6 CS 53 Lectre 2 Virtal Memory VP 7 8

9 Handling Page Falt (2) Page falt handler selects a victim to be evicted (here VP 4) Virtal address PTE PTE 7 Valid Physical page nmber or disk address nll nll Memory resident page table Physical memory VP VP 7 VP 4 Virtal memory (disk) VP VP 3 VP 4 VP 6 PP PP 3 CS 53 Lectre 2 Virtal Memory VP 7 9

10 Handling Page Falt (3) Evict the content of VP4 Virtal address PTE PTE 7 Valid Physical page nmber or disk address nll nll Memory resident page table Physical memory VP VP 7 VP 4 Virtal memory (disk) VP VP 3 PP PP 3 VP 4 VP 6 CS 53 Lectre 2 Virtal Memory VP 7

11 Handling Page Falt (4) Update page table Virtal address Valid PTE PTE 7 Physical page nmber or disk address nll nll Memory resident page table Physical memory VP VP 7 Empty Virtal memory (disk) VP VP 3 PP PP 3 VP 4 VP 6 CS 53 Lectre 2 Virtal Memory VP 7

12 Handling Page Falt (5) Load content of VP3 to DRAM Virtal address PTE PTE 7 Valid Physical page nmber or disk address nll nll Memory resident page table Physical memory VP VP 7 VP 3 Virtal memory (disk) VP VP 3 PP PP 3 VP 4 VP 6 CS 53 Lectre 2 Virtal Memory VP 7 2

13 Handling Page Falt (6) Update page table Virtal address Valid PTE PTE 7 Physical page nmber or disk address nll nll Memory resident page table Physical memory VP VP 7 VP 3 Virtal memory (disk) VP VP 3 PP PP 3 VP 4 VP 6 CS 53 Lectre 2 Virtal Memory VP 7 3

14 Handling Page Falt (7) Restart the instrction: page hit! Virtal address PTE PTE 7 Valid Physical page nmber or disk address nll nll Memory resident page table Physical memory VP VP 7 VP 3 Virtal memory (disk) VP VP 3 PP PP 3 VP 4 VP 6 CS 53 Lectre 2 Virtal Memory VP 7 4

15 Today Virtal memory Memory as cache Page/cache replacement policy CS 53 Lectre 2 Virtal Memory 5

16 Page Replacement When a page falt occrs, the OS loads the falted page from disk into a page frame of memory At some point, the process has sed all of the page frames it is allowed to se This is likely (mch) less than all of available memory When this happens, the OS mst replace a page for each page falted in It mst evict a page to free p a page frame Written back only if it is has been modified (i.e., dirty )! CS 53 Lectre 2 Virtal Memory 6

17 Page replacement policy What we discssed so far (page falts, swap, page table strctres, etc ) is mechanisms Page replacement policy: determine which page to remove when we need a victim Does it matter? Yes! Page falts are sper expensive Getting the nmber down, can improve the performance of the system significantly CS 53 Lectre 2 Virtal Memory 7

18 Considerations Page replacement spport has to be simple dring memory accesses They happen all the time, we cannot make that part slow Bt it can be complicated/expensive when a page falt occrs why? Reason : if we are sccessfl, this will be rare Reason 2: when it happens we are paying the cost of I/O» I/O is very slow: can afford to do some extra comptation» Worth it if we can save some ftre page falts What makes a good page replacement policy? CS 53 Lectre 2 Virtal Memory 8

19 Locality to the Resce Recall that virtal memory works becase of locality Temporal and spatial Work at different scales: for cache, at a line level, for VM, at page level, and even at larger scales All paging schemes depend on locality What happens if a program does not have locality? High cost of paging is acceptable, if infreqent Processes sally reference pages in localized patterns, making paging practical CS 53 Lectre 2 Virtal Memory 9

20 Evicting the Best Page Goal is to redce the page falt rate The best page to evict is the one never toched again Will never falt on it Never is a long time, so picking the page closest to never is the next best thing Evicting the page that won t be sed for the longest period of time minimizes the nmber of page falts Proved by Belady We re now going to srvey varios replacement algorithms, starting with Belady s CS 53 Lectre 2 Virtal Memory 2

21 Belady s Algorithm Belady s algorithm Idea: Replace the page that will not be sed for the longest time in the ftre Optimal? How wold yo show? Problem: Have to predict the ftre Why is Belady s sefl then? Use it as a yardstick/pper bond Compare implementations of page replacement algorithms with the optimal to gage room for improvement» If optimal is not mch better, then algorithm is pretty good What s a good lower bond?» Random replacement is often the lower bond CS 53 Lectre 2 Virtal Memory 2

22 First-In First-Ot (FIFO) FIFO is an obvios algorithm and simple to implement Maintain a list of pages in order in which they were paged in On replacement, evict the one broght in longest time ago Why might this be good? Maybe the one broght in the longest ago is not being sed Why might this be bad? Then again, maybe it s not We don t have any info to say one way or the other FIFO sffers from Belady s Anomaly The falt rate might actally increase when the algorithm is given more memory (very bad) CS 53 Lectre 2 Virtal Memory 22

23 Least Recently Used (LRU) LRU ses reference information to make a more informed replacement decision Idea: We can t predict the ftre, bt we can make a gess based pon past experience On replacement, evict the page that has not been sed for the longest time in the past (Belady s: ftre) When does LRU do well? When does LRU do poorly? Implementation To be perfect, need to time stamp every reference (or maintain a stack) mch too costly So we need to approximate it CS 53 Lectre 2 Virtal Memory 23

24 Approximating LRU LRU approximations se the PTE reference bit Keep a conter for each page At reglar intervals, for every page do:» If ref bit =, increment conter» If ref bit =, zero the conter» Zero the reference bit The conter will contain the nmber of intervals since the last reference to the page The page with the largest conter is the least recently sed Some architectres don t have a reference bit Can simlate reference bit sing the valid bit to indce falts CS 53 Lectre 2 Virtal Memory 24

25 LRU Clock (Not Recently Used) Not Recently Used (NRU) Used by Unix Replace page that is old enogh Arrange all of physical page frames in a big circle (clock) A clock hand is sed to select a good LRU candidate» Sweep throgh the pages in circlar order like a clock» If the ref bit is off, it hasn t been sed recently What is the minimm age if ref bit is off?» If the ref bit is on, trn it off and go to next page Arm moves qickly when pages are needed Low overhead when plenty of memory If memory is large, accracy of information degrades» What does it degrade to?» One fix: se two hands (leading erase hand, trailing select hand) CS 53 Lectre 2 Virtal Memory 25

26 LRU Clock P3: P3: P2: P4: P2: P4: P: P5: P: P5: P8: P6: P8: P6: P7: P7: CS 53 Lectre 2 Virtal Memory 26

27 Example: gcc Page Replace CS 53 Lectre 2 Virtal Memory 27

28 Example: Belady s Anomaly CS 53 Lectre 2 Virtal Memory 28

29 Other ideas Victim bffer Add a bffer (death row!) we pt a page on when we decide to replace it Bffer is FIFO If yo get accessed while on death row clemency! If yo are the oldest page on death row replacement! CS 53 Lectre 2 Virtal Memory 29

30 Fixed vs. Variable Space In a mltiprogramming system, we need a way to allocate memory to competing processes Problem: How to determine how mch memory to give to each process? Fixed space algorithms» Each process is given a limit of pages it can se» When it reaches the limit, it replaces from its own pages» Local replacement Some processes may do well while others sffer Variable space algorithms» Process set of pages grows and shrinks dynamically» Global replacement One process can rin it for the rest CS 53 Lectre 2 Virtal Memory 3

31 Working Set Model A working set of a process is sed to model the dynamic locality of its memory sage Defined by Peter Denning in 6s Definition WS(t,w) = {set of pages P, sch that every page in P was referenced in the time interval (t, t-w)} t time, w working set window (measred in page refs) A page is in the working set (WS) only if it was referenced in the last w references CS 53 Lectre 2 Virtal Memory 3

32 Working Set Size The working set size is the nmber of pages in the working set The nmber of pages referenced in the interval (t, t-w) The working set size changes with program locality Dring periods of poor locality, yo reference more pages Within that period of time, the working set size is larger Intitively, want the working set to be the set of pages a process needs in memory to prevent heavy falting Each process has a parameter w that determines a working set with few falts Denning: Don t rn a process nless working set is in memory CS 53 Lectre 2 Virtal Memory 32

33 Example: gcc Working Set CS 53 Lectre 2 Virtal Memory 33

34 Working Set Problems Problems How do we determine w? How do we know when the working set changes? Too hard to answer So, working set is not sed in practice as a page replacement algorithm However, it is still sed as an abstraction The intition is still valid When people ask, How mch memory does Firefox need?, they are in effect asking for the size of Firefox s working set CS 53 Lectre 2 Virtal Memory 34

35 Page Falt Freqency (PFF) Page Falt Freqency (PFF) is a variable space algorithm that ses a more ad-hoc approach Monitor the falt rate for each process If the falt rate is above a high threshold, give it more memory» So that it falts less» Bt not always (FIFO, Belady s Anomaly) If the falt rate is below a low threshold, take away memory» Shold falt more» Bt not always Hard to se PFF to distingish between changes in locality and changes in size of working set CS 53 Lectre 2 Virtal Memory 35

36 Thrashing Page replacement algorithms avoid thrashing When most of the time is spent by the OS in paging data back and forth from disk No time spent doing sefl work (making progress) In this sitation, the system is overcommitted» No idea which pages shold be in memory to redce falts» Cold jst be that there isn t enogh physical memory for all of the processes in the system» Ex: Rnning Windows95 with 4 MB of memory Possible soltions» Swapping write ot all pages of a process» By more memory CS 53 Lectre 2 Virtal Memory 36

37 Smmary Page replacement algorithms Belady s optimal replacement (minimm # of falts) FIFO replace page loaded frthest in past LRU replace page referenced frthest in past» Approximate sing PTE reference bit LRU Clock replace page that is old enogh Working Set keep the set of pages in memory that has minimal falt rate (the working set ) Page Falt Freqency grow/shrink page set as a fnction of falt rate Mltiprogramming Shold a process replace its own page, or that of another? CS 53 Lectre 2 Virtal Memory 37

38 Next time Disk Drives Preparation Read Modle 35, 36, 37, 44 CS53 Lectre 6 Paging 38