課程名稱︰作業系統
課程性質︰必修
課程教師︰薛智文
開課學院：電資學院
開課系所︰資工系
考試日期（年月日）︰2019/06/19
考試時限（分鐘）：180
試題 :
This is an open-book and open-own-note exam. Please do answer with your own
words in the ORDER of question number. Provide your own question definition if
you feel it is not clear enough. Points will be deducted if your answer is
difficult to read. You can write in Chinese and keep this paper. Have a nice
summer vacation.
1. [10%] Suppose there is no deadlock prevention and detection implemented in a
computer system. How do we make sure there is a deadlock happening? [5%] How
do we find the processes involved in the deadlock? [5%]
2. [10%] Why do we need different sizes of a memory page? [5%] If we support
different page sizes at the same time in an OS, what need to be changed in
TLB? [5%]
3. [10%] Why we used to lock shared pages in memory? [5%]? For locking IO
memory pages, is there any different reason? [5%].
4. [10%] When implementing an application, we usually experiment in higher
(application) level for flexibility or avoiding system crash, then move into
lower (kernel, driver or even hardware) level for better performance. In
what sample applications, we would tune in higher level for better
performance [5%] and lower level for flexibility, respectively. [5%]
5. [10%] Any way to improve the design of working set for lowering the page
fault rate? [5%] Why there are upper bound and lower bound to control page
fault frequency? [5%]
6. [10%] Describe the key factors to make page fault a successful design to
virtual memory management utilizing the secondary storage. [5%] When the
space of secondary storage such as disk is not enough, how can we further
use the tertiary storage such as tape? [5%]
◎ 黑板補充： tertiary 第三
7. [20%] Memory management and (disk) storage management use similar
technologies.
a. List 3 common ones. [1,2,2%]
b. Describe 3 technologies which are better used in memory management but
not in storage management, with the most significant reason in one
sentence. [3*5%] excluding page fault.
8. [20%] For Project 2, please answer the following questions:
a. Since it is a client-server program, how do we guarantee the correct
behavior no matter how the set up orders of both sides? [5%]
b. Where enforcing "Copy Semantics" might be beneficial in your design? [5%]
c. Please describe the performance differences of different file sizes,
buffer sizes, and implementations (synchronous or asynchoronous, ...).
[10%]
8. [20%] For a multiprogramming system with many process, when a process needs
to do I/O operations, it will send a system call to the operating system,
but if the controller is busy, the process will be placed in the disk queue,
allowing the operating system to execute or optimize later. There are some
common disk scheduling algorithms: FCFS, SSTF, SCAN, C-SCAN, LOOK, C-LOOK
a. Suppose there are totally of 200 cylinders on the disk, if it requests
continuous read and write track 105, 45, 120, 130, 70, 111, and 155.
Which scheduling algorithm results in the least disk arm moves, and how
many? (Head start on the 50th track, the default direction is the
direction of increasing) [7%]
b. When the probability to read any track varies little (probability needs
not to be fixed on each track), try to analyze the performance of the
best and the worst scheduling algorithm? [4%]
c. What applications are mostly suitable using FCFS, C-SCAN, SSTF,
respectively? Why? [3*3%]