1. Avishai Wool lecture 8 - 1 Introduction to Systems Programming Lecture 8 Paging Design

2. Avishai Wool lecture 8 - 2 Steps in Handling a Page Fault

3. Avishai Wool lecture 8 - 3 Virtual  Physical mapping CPU accesses virtual address 100000 MMU looks in page table to find physical address –Page table is in memory too Unreasonable overhead!

4. Avishai Wool lecture 8 - 4 TLB: Translation Lookaside Buffer Idea: Keep the most frequently used parts of the page table in a cache, inside the MMU chip. TLB holds a small number of page table entries: Usually 8 – 64 TLB hit rate very high because, e.g., instructions fetched sequentially.

5. Avishai Wool lecture 8 - 5 A TLB to speed up paging Example: Code loops through pages 19,20,21 Uses data array in pages 129,130,140 Stack variables in pages 860,861

6. Avishai Wool lecture 8 - 6 Valid TLB Entries TLB miss: –Do regular page lookup –Evict a TLB entry and store the new TLB entry –Miniature paging system, done in hardware When OS does context switch to a new process, all TLB entries become invalid: –Early instructions of new process will cause TLB misses.

7. Avishai Wool lecture 8 - 7 TLB placement/eviction Done by hardware Placement rule: –TLBIndex = VirtualAddr modulo TLBSize –TLBSize is always 2 k  TLBIndex = k least-significant bits –Keep “tag” (rest of bits) to fully identify virtual addr Virtual address can be only in one TLB index No explicit “eviction”: simply overwrite what is in TLB[TLBIndex]

8. Avishai Wool lecture 8 - 8 TLB + Page table lookup In TLB? In page table? Page fault: copy from disk to memory Virtual address Physical address No Yes Yes; update TLB No

9. Avishai Wool lecture 8 - 9 TLB – cont. If address is in TLB  page is in physical memory –OS invalidates TLB entry when evicting a page –So page fault not possible if we have a TLB hit “page fault rate” is computed only on TLB misses

10. Avishai Wool lecture 8 - 10 Example: Average memory access time TLB lookup: 4ns Phys mem access: 10ns Disk access: 10ms TLB miss rate: 1% Page fault rate: 0.1% Assume page table is in memory. p=0.99, time=4ns+10ns Page hit: p=0.01*0.999, time=4ns+10ns+10ns Page fault: p=0.01*0.001, time=4ns+10ns+10ms+10ns TLB miss TLB hit Average memory access: 114.1ns (1.141*10 -7 )

11. Avishai Wool lecture 8 - 11 Design issues in Paging

12. Avishai Wool lecture 8 - 12 Local versus Global Allocation Policies: Physical Memory a)Original configuration – ‘A’ causes page fault b)Local page replacement c)Global page replacement

13. Avishai Wool lecture 8 - 13 Local or Global? Local  number of frames per process is fixed –If working set grows  thrashing –If working set shrinks  waste Global usually better Some algorithms can only be local (working set, WSClock).

14. Avishai Wool lecture 8 - 14 How many frames to give a process? Fixed number Proportional to its size (before load) Zero, let it issue page faults for all its pages. –This is called pure demand paging. Monitor page-fault-frequency (PFF), give more pages if PFF high.

15. Avishai Wool lecture 8 - 15 Page fault rate as a function of the number of page frames assigned

16. Avishai Wool lecture 8 - 16 Load Control Despite good designs, system may still thrash When PFF algorithm indicates –some processes need more memory –but no processes need less Solution: Reduce number of processes competing for memory –swap one or more to disk, divide up frames they held –reconsider degree of multiprogramming

17. Avishai Wool lecture 8 - 17 Cleaning Policy Need for a background process, paging daemon –periodically inspects state of memory When too few frames are free –selects pages to evict using a replacement algorithm It can use same circular list (clock) –as regular page replacement algorithm but with diff ptr

18. Avishai Wool lecture 8 - 18 Windows XP Page Replacement Processes are assigned working set minimum and working set maximum Working set minimum is the minimum number of page frames the process is guaranteed to have in memory A process may be assigned as many page frames up to its working set maximum When the amount of free memory in the system falls below a threshold, automatic working set trimming is performed to restore the amount of free memory Working set trimming removes frames from processes that have more than their working set minimum