Download CS162 Midterm Exam 2 Solutions - OS, Virtual Memory, File Buffer Cache and more Exams Operating Systems in PDF only on Docsity!
University of California, Berkeley College of Engineering Computer Science Division – EECS
Fall 1999 Anthony D. Joseph
Midterm Exam #2 Solutions
November 10, 1999 CS162 Operating Systems
Your Name:
SID:
Circle the letters of your CS162 login (1 per line):
a b c d e f g h i j k l m n o p q r s t u v w x y z
a b c d e f g h i j k l m n o p q r s t u v w x y z
TA Name / Section:
This is a closed book examination. You have two hours to answer as many questions as possible. The number in parentheses at the beginning of each question indicates the number of points given to the question; there are 100 points in all. You should read all of the questions before starting the exam, as some of the questions are substantially more time consuming.
Write all of your answers directly on this paper. Make your answers as concise as possible. If there is something in a question that you believe is open to interpretation, then please ask us about it!
Good Luck!!
Problem Possible Score 1 12
2^28
3^15
4^12
Total 100
Solutions
- (12 points total) True / False and Why. Circle the correct answer and provide a short reason for your answer: a. Doubling the block size in the UNIX 4.2 BSD file system will exactly double the maximum file size.
True / False
Why: False. The size will more than double because the size of indirect, doubly indirect, and triply indirect blocks also doubles.
The correct answer was worth 2 points. Good reasoning was worth 2 points.
b. A direct mapped cache can sometimes have a higher hit rate than a fully associative cache with an LRU replacement policy (on the same reference pattern).
True / False
Why: True. A direct mapped cache will do better than LRU on a pattern like ABCDEABCDE… if the cache size is one entry smaller than the total number of items in the pattern (e.g., four cache entries). LRU will miss on every access, while a direct mapped cache will only miss on the two entries that map to the same cache entry.
c. Virtual memory address translation is useful even if the total size of virtual memory (summed over all programs) is guaranteed to be smaller than physical memory.
True / False
Why: True. It avoids the need to relocate programs when they are loaded and allows memory to be dynamically allocated. Also, provides protection and doesn’t require that each program’s addresses be contiguous.
Solutions
b. (14 points) Specify how you implement a write-behind policy. Assume a write- behind occurs every 30 seconds. Assume that a file is only opened by one process at a time. You do not have to show how you would modify the read operation.
i) List the data structures, methods, or classes you would either use or modify:
Use a global SyncList, writesPending, (or alternatively, per-process SyncLists) of ToWrite structures, containing : the ID of the open file, a buffer for holding the data to be written, and an offset into the file (where the data is to be written). We graded solutions to (i) and (ii) by subtracting points for errors:
- Major errors subtracted 4 points each, up to –12 total. Examples: using a thread per request (has a severe performance impact and could result in out-of-order writes), missing locks/synchronization, explicitly using interrupts to perform write-behind (can’t do I/O in an interrupt handler).
- Minor errors cost 2 points each, up to -6 total. Example, not defining data structures, or failing to say how the kernel procedure for write-behind was executed (many solutions relied on the procedure being called by an interrupt handler). If interrupts were not explicitly used, we only counted this as a minor error.
- For solutions slightly over the length limit, we subtracted 2 points; for major overruns, 4 points; for extremely long solutions, 8 points.
ii) Provide the algorithm for the policy in pseudocode form. Your solution should take less than twenty lines of pseudocode. No optimizations are necessary, however your solution should handle any synchronization issues. You can use any Nachos functions that you would need (if they’re not standard, or are from your own project phases, explain their operation).
Create a new kernel thread for write-behind that does the following: while (1) { WaitUntil(30 seconds); while (writesPending is not empty) { dequeue ToWrite element write buffer to file free ToWrite element } }
Add to write procedure: create ToWrite element copy user data to buffer fill in element fields append to writesPending
Solutions
(This page intentionally left blank)
Solutions
b. (5 points) The HAL Junior price sheet looks like this:
Item Specs Price Larger TLB Reduces the probability of a TLB miss to 0.05. Assume that there is sufficient memory that the page fault rate is unaffected.
Hard Disk-Drive II Transfers a page in 8 milliseconds $ 8 MByte more memory
Makes probability of a page fault, given a TLB miss, 0.
Suppose your friend has $1000. Which components should they buy if they want to maximize the performance of their HAL Junior? State the reasoning behind your choices.
TLB => 1 + 0.05 * 4 => 1. HDII => 1 + 0.1 * 3.4 => 1. Mem => 1 + 0.1 * (1 + 0.0001 * 15,000) => 1+ 0.1 * 2.5 => 1. TLB + Mem => 1. TLB + HDII => 1. HDII + Mem => 1.
All are good choices, but the best choice is to spend $500 on TLB and $500 on memory.
Each correct choice was worth 2 points. A good argument was worth one point.
Solutions
No Credit – Problem X: (000000000000 points)
The news is not very good for Bill Gates. On Friday, the judge in the antitrust trial of Microsoft ruled that the software company has monopoly power in PC operating systems and that it has used that power to crush potential threats from competitors.
The Top 10 Things on Bill Gates's To-Do List
- Change menacing cackle to more of a charming titter.
- Stop payment on Satan's check.
- Search JobOptions.com: Field = "Technology" Salary > $25 Billion
- Put somebody else in charge temporarily; take the winter off and find Rosebud.
- Push own "Start" button. At prompt, choose "Shut Down" and then "Re-start Ego."
- Create new corporate division in charge of sending flowers and candy to Sandy O'Connor.
- Send message to mother ship: "My job here is done."
- Dedicate my life to finding the real monopolists.
- See how quickly the government can prepare for the "11/10/99 Bug."
- Halt global economy by taking all my money and going home.
Solutions
- (15 points total) Virtual memory page replacement policies. You are given a program that repeatedly scans through the elements of a very large array in virtual memory. For example, if the array is 4 pages long, then its page reference pattern is ABCDABCDABCD...
For each of the following page replacement algorithms, sketch a graph showing the paging behavior. Assume that the program has been running for several iterations through the array. The y-axis of each graph is the number of page faults per referenced page , varying from 0 to 1; the x-axis is the size of the array being scanned, varying from smaller than physical memory to much larger than physical memory. Label any interesting points on the graph on both the x and y axes.
a. FIFO
Size of array
Number of page faults per referenced page
Size of physical memory
b. LRU
Size of array
Number of page faults per referenced page
Size of physical memory
Solutions
c. MIN
Size of array
Number of page faults per referenced page
Size of physical memory
Asymptotic to 1
For FIFO and LRU, we subtracted two points if there was a significant slope to transition from 0 to 1. We also subtracted two points if the graph did not have a zero fault rate before the size of physical memory was reached (we specifically state that the program has been running for some time).
For MIN, we subtracted two points if the graph was not asymptotic to 1 or if it did not have a zero fault rate before the size of physical memory was reached.
