Computer Science 432 |
Lab 6: Working Set Simulator
Due: 9:00 AM, Monday, November 13, 2006
Consider the following program segment, written in a C-like language:
const int n=10; int i, j, A[n], B[n], C[n], temp; for (i=1; i<=n; i++) { A[i]=i; B[i]=n-i+1; } for (i=1; i<=n; i++) { temp=0; for (j=i; j<=n; j++) { temp=temp+A[n+i-j]*B[j]; } C[i]=temp; }
Using a machine with registers denoted by Ri and a fixed instruction size of 1 word per instruction, the machine language version of this program is loaded in virtual address space (with page size 4K, i.e., 1024 words) as follows:
0x2FBC (R1) <- ONE Index i 0x2FC0 (R2) <- n Loop bound 0x2FC4 compare R1,R2 Test i>n 0x2FC8 branch_greater * + 0x20 0x2FCC A(R1) <- (R1) Compute A[i] 0x2FD0 (R0) <- n Compute B[i] 0x2FD4 (R0) <- (R0) - (R1) 0x2FD8 (R0) <- (R0) + ONE 0x2FDC B(R1) <- (R0) 0x2FE0 (R1) <- (R1) + ONE Increment i 0x2FE4 branch * - 0x20 0x2FE8 (R1) <- ONE Index i 0x2FEC (R2) <- n Loop bound 0x2FF0 compare R1,R2 Test i>n 0x2FF4 branch_greater * + 0x50 0x2FF8 (R0) <- ZERO temp <- 0 0x2FFC temp <- (R0) 0x3000 (R3) <- (R1) Index j 0x3004 (R4) <- n Loop bound 0x3008 compare R3,R4 Test j>n 0x300C branch_greater * + 0x20 0x3010 (R0) <- n Compute A[n+i-j] 0x3014 (R0) <- (R0) + (R1) 0x3018 (R0) <- (R0) - (R3) 0x301C (R5) <- A(R0) 0x3020 (R6) <- B(R3) Compute B[j] 0x3024 (R5) <- (R5) * (R6) 0x3028 (R5) <- (R5) + temp 0x302C temp <- (R5) 0x3030 (R3) <- (R3) + ONE Increment j 0x3034 branch * - 0x20 0x3038 C(R1) <- (R5) Compute C[i] 0x303C (R1) <- (R1) + ONE Increment i 0x3040 branch * - 0x50 ... 0x6000 Storage for C 0x7000 Storage for ONE 0x7004 Storage for n 0x7008 Storage for temp 0x700C Storage for ZERO 0x8000 Storage for A 0x9000 Storage for B
Upon execution of this program segment, the following reference string is generated:
omega = 272722(28272272927222)n272722(272733733(373338393373737333)n-i+13637322)nWrite a program (in the language of your choice) that simulates (yes, another simulator) the run-time behavior of this program segment when a working set memory management policy is used. The program should print values:
Delta | = | window size |
P(Delta) | = | total number of page faults |
W(Delta) | = | average working set size |
F(Delta) = (P(Delta))/(|omega|) | = | average page fault rate |
Note that as each entry in the reference string (page) is processed, one of four things will happen to the working set. (i) the page is added to the set, and none is removed, (ii) the page is added to the set and one old page is removed, (iii) the page is already in the set and another page is removed, or (iv) the page is already in the set and no other page is removed.
Your program may hard-code the reference string (or generate it on the fly) but at least make things modular enough that you could replace one function to generate a different reference string.
If you do not get your simulator working in time, you may use my program to generate your plots and answer the questions.
You may work individually or in groups of two or three. Submit your program as a tar file lab6.tar. Make sure you include a Makefile, a README file that includes build instructions if it is anything beyond "type make", and a PDF file lab6.pdf that includes your plots and answers to the questions.