Part 1 – Nice (20 Points)

Having policies is the first step in implementing a non-trivial scheduler. For the sake of simplicity, you can implement the nice UNIX API as an extra system call (in xv6). A higher nice value means lower priority. For more information type – man 2 nice on Linux system.

You should be able to store the nice value in the process structure. Xv6 stores the PCB in struct proc in the file proc.h

Note: You are not expected to enforce specific permissions on nice. On Linux systems, typically only root can increase priorities. But you can skip this part in the assignment.

Implement the nice system call in xv6. Explain your implementation in the PDF in 1~2 paragraphs. To verify that the nice value can be set and read back correctly you would need to write a few test cases. Your test cases must include the ability to handle edge cases. What happens if the values passed in are out of bounds? Provide screenshot(s) of your test cases along with an explanation. Also, mention how to run your test cases – in what file is it located and what to run.

Part 2 – Random Number Generator (20 Points)

Lottery scheduling is a randomized heuristic. Hence, we need to design a pseudo- random number generator (PRNG). Since xv6 does not include a PRNG, you need to design one. This does not require strong randomness such as in cryptographic algorithms but a decent PRNG will do.

We recommend using the Xorshift family, although you can use anything else. If you do choose to use the Xorshift values, care should be taken to pick non-zero seed values.

Implement an in-kernel PRNG and write a simple test case to verify that the values returned look evenly distributed. Explain your implementation in the PDF in 1~2 paragraphs. Provide screenshot(s) of your test cases along with an explanation. Also, mention how to run your test cases – in what file is it located and what to run.

Hints:

– For the purpose of this assignment, to test that your PRNG function works, you could run it 1000 times, and test for the max, min, and mean to see if the values are relatively evenly distributed and that the numbers are not heavily skewed. You may plot a histogram to check your even distribution and include it in your report.

Part 3 – Scheduler (60 Points)

Mentioned in the link, lottery scheduling works by assigning tickets to each of the processes in the system. Then, in each time slice, it randomly picks a ‘winning’ ticket. It is up to you to decide how many more tickets to assign less nice processes and vice versa.

The design problem is more open-ended. After reading the text mentioned in the link above, some of the design factors you might want to consider:

– Which data structure makes sense?

– In what way would you handle a process leaving or entering a system all while

maintaining a policy?

– How would you map the winning number back to the runnable process?

Note: You should keep both the schedulers as an option. It is fine to select the scheduler (Round Robin vs. Lottery) to be a compile time option.

Implement lottery scheduling in xv6. As mentioned, the particulars of the data structure and the policies to map the number of tickets to nice values is a design choice which is open-ended. You have to convince us that your implementation handles edge cases. Provide a detailed explanation of your approach. (Questions like ‘what if?’ should be mentioned and handled as well)

Come up with at least 3 test cases for your lottery scheduler and provide screenshot(s) of your test cases along with an explanation. Also, mention how to run your test cases – in what file is it located and what to run.

Hints:

– You can fork several children doing the same task with different priorities. Each of the children can print the progress in incremental points.

– You may change the priority of a process dynamically and check if its progress accelerates or slows down appropriately.