CS-350 - Homework Assignment #1 – EVAL

EVAL Problem 1

In this EVAL problem, you will evaluate the result of the clock measuring functions developed in the corre-sponding BUILD assignment. In particular, we will estimate which method (SLEEP vs. BUSYWAIT) works best for stable CPU speed estimates, and whether the resulting measurements match the CPU speci cations.

    a) First, estimate the CPU clock speed on your own machine using the SLEEP method. To do this, collect the results returned by the clock executable you implemented by running it 10 times. Start with a wait time of 1 second, all the way up to (and including) 10 seconds, increasing the wait time by 1 second at each step.

From the 10 CPU clock speed measurements you have obtained, compute and report the max, min, average, and standard deviation.

    b) Repeat the same procedure above where instead of the SLEEP method, you use the BUSYWAIT method for all the measurements. Then, compute and report the same statistics on the measured CPU clock speed.

    c) Based on what you have acquired in the previous steps, compare the SLEEP and BUSYWAIT methods to estimate the CPU speed of your processor. Which method yielded more precise and/or stable results? Take also a look at the speci cations of your CPU. Which method produced a result that is closer to the speci cations provided by the manufacturer of your CPU? Comment on whether the results are in a match with said speci cations or not, and why.








































2
CS-350 - Fundamentals of Computing Systems::Homework Assignment #1 - EVAL    Problem 2


EVAL Problem 2

In this EVAL problem, we will start to examine the behavior of the server process in response to variable tra c conditions generated by the client.

HINT: many of the questions below, will ask you to determine a lapse of time in which the server was active. You can compute that time window in post-processing as the lapse of time between the <receipt timestamp> of the very rst request and the <completion timestamp> of the last request processed by the server.

    a) First, launch the client with the following parameters: client -a 10 -s 12 -n 500 <some port> where <some port> is the same port (of your choice) used for the server. Allow the client and server to exchange data and terminate, then post-process the output produced by the server.

From the server output, compute the average throughput of the server in terms of requests per second. Walk us through the steps you performed to accomplish this.

    b) Follow the same procedure above and once again post-process the output of the server. This time, compute the utilization of the server as a percentage. Walk us through the steps you performed to accomplish this.

    c) Repeat the experiment above but by running the server and client back-to-back, while at the same time extracting some statistics from the server run from the OS. To do so, move to the build directory and use the following command line:

/usr/bin/time -v ./server 2222 & ./client -a 10 -s 12 -n 500 2222

NOTE: by doing so, BOTH client and server will run in the same terminal and their output will be interleaved on the console. If you want to suppress the output of the client, redirect the stdout of the client to the null device using:

/usr/bin/time -v ./server 2222 & ./client -a 10 -s 12 -n 500 2222 > /dev/null

Also note that the server will run in background. You can use the following command to kill it if it does not terminate:

killall server

After the server has run successfully, the time utility will print a host of statistics acquired from the OS. Take a look at the entry \Percent of CPU this job got:". Does it match with what you computed in the previous question?

    d) Now let us repeat the computation of the utilization of the server as in Qb) but this time sweep through the -a parameter passed to the server. In particular, run the rst experiment for a value of 1; then a second time with a value of 2; and so on until and including the case where the value is 12. Thus, you will run 12 experiments in total.

Produce a plot that depicts the trend of the resulting server utilization (y-axis) as a function of the arrival rate (-a) parameter (x-axis). Is there any correlation between the two values?

    e) By reusing the same exact output les produced as part of the previous question, let us reason about the response time of the various requests. First, compute the average response time for all the 500 requests as observed in the case when the parameter -a was set to 10. Also compute max, min, and std. deviation.

    f) Now, repeat the average response time calculation individually for the various runs with the -a pa-rameter from 1 to 12. Finally, produce a plot that depicts the trend of the average response time y-axis as a function of the server utilization x-axis. What relationship do you discover between the two quantities?



3