Test 2
Operating Systems and Networking
CS330
Spring 2007
Answer only five from the following
Allotted time: 1 hr and 50 minutes
Date: March 14, 2007
Policy: Open notes, text and reference books.
1. CPU periodically receives data from devices like keyboard, mouse, network interface card, etc. To
figure out if there’s any data from a device for CPU, it must employ some method like polling or
interrupt from the device to CPU. Briefly discuss when one is more efficient than the other.
Polling is CPU intensive, but simple. If CPU has nothing else to do, it may not be a major
problem. In a multitasking, it’s expensive to devote precious CPU time on polling pursuits. In
interrupt driven I/O, one needs extra hardware. The situation becomes complicated when the
system is connected to multiple interrupt lines to receive interrupt from multiple devices. In such
events polling would be desirable. If we know in advance polling would not cause too much of
CPU time drainage, polling would be preferred. If we know a priori that the likelihood of an
interrupt from a device is low, interrupt would be the chosen scheme.
2. For a system of interest, assume the context switching time in a RR system is negligible. Each
process when it arrives at CPU gets
units of time-slice. Will I/O utilization increase or
decrease if
is increased? State your rationale.
On a pure CPU-bound or IO-bound jobs, I/O utilization will not change. However, on a mixed
stream I/O utilization will decrease. Since I/O bound processors, on arrival in CPU, request for I/O
processing density of such requests will decline with the increasing size of
. In this mode, the
CPU-bound jobs tend to hang around the CPU longer delaying the eventual arrival time of I/O
bound processes at CPU. Accordingly, I/O processors would be forced to wait resulting in a
decrease of I/O device utilization.
3. Consider the same system as outlined in the previous question. If
is reduced, will the reduction
cause some programs to take longer times to finish? Why, or why not?
For some programs, it will take longer. Suppose there are programs requiring exactly 1 time-slice.
If
is reduced, they will take 2 or more time-slices to complete. Also, it would allow some
programs to finish earlier if they needed less than one CPU time slice currently being used.
4. In the lecture, it was pointed out that a busy-waiting by a process to get into its critical section
should generally be avoided, if possible. Indicate why. Could there be any situation, however,
where busy-waiting might not be all that bad? Indicate how.
If busy-waiting is likely to involve wasting only a small fragment of CPU time on the average,
then it is definitely worth it. Alternatives to busy-waiting via semaphore or monitor module entail
a great deal of overhead cost.
5. The two processes shown below access a shared printer to do their print tasks via a semaphore S
which is initialized to 0. It is conjectured that the following output strings are either individually
valid or invalid when the two processes are run concurrently. Mark the output strings either as
valid or invalid strings.
