The final project of the course regards the implementation of a User Mode thread Scheduling, inspired from the UMS implementation available in the Windows kernel. The project is intended to be developed individually, or in a group of maximum 2 people by means of GitHub classroom.
You are requested to implement a User Mode thread Scheduling (UMS) facility which makes able the programmer of a C user space application to schedule threads without involving the kernel scheduler. The ability to switch threads in user mode makes UMS more efficient than thread pools for managing large numbers of short-duration work items that require few system calls.
The application which is intended to use UMS needs to implement a UMS scheduler component which is in charge to determine which is the thread to be scheduled. The creation of the scheduler receives as input:
- an entrypoint, that is a function that is executed for determining the next thread to be scheduled and therefore it is called when a the scheduled is started the first time, when a work thread ends, yields (with
UmsThreadYield, see below) or is put on wait (e.g. page fault);
- a set of worker threads, that we also call a completion list.
Multiple UMS schedulers can share threads or completion lists (so be careful for race conditions), and a UMS Scheduler has always associated a scheduler thread (in general an application creates a scheduler thread for each core/processor available in the system). The scheduler thread will host the execution of the worker threads.
You, at least, have to implement:
EnterUmsSchedulingModewhich converts a standard pthread in a UMS Scheduler thread, the function takes as input a completion list of worker threads and a entry point function;
DequeueUmsCompletionListItemscalled from the scheduler thread obtains a list of current available thread to be run, if no thread is available to be run the function should be blocking until a thread becomes available;
ExecuteUmsThreadcalled from a scheduler thread, it executes the passed worker thread by switching the entire context;
UmsThreadYieldcalled from a worker thread, it pauses the execution of the current thread and the UMS scheduler entry point is executed for determining the next thread to be scheduled;
The backend implementation is up to you, you can also decide to build a worker thread ready queue (as in Windows specification) in order to speed up the lookup of ready threads. In that case, implement also the
GetNextUmsListItem that adds a worker thread from the list returned by
DequeueUmsCompletionListItems to the ready queue of the scheduler.
You are also required to implement:
- expose in
/proc/ums/<pid>/schedulersa folder for each scheduler thread (named with numbers from 0 to n), and inside that folder a file named
info, that shows some statistics as the number of times the scheduler switched to a worker thread, the completion list, the time needed for the last worker thread switch, the current state (idle/running) and if it is running, which worker thread in the completion list;
- expose in
/proc/ums/<pid>/schedulers/<id>/workersa file for each worker thread (named with numbers from 0 to n as in the sorting order of the completion list) that shows some statistics, as the total running time of the thread, the number of switches, the current state (idle/running)
You can and need to implement any other function you think that is useful for adhering to track.
Every notion, detail, particular behaviour, parameter or variable that is not mentioned in this track is left to you to decide and if you have doubt you can take inspiration from the original Windows specification (see References). You do not need to be fully compatible with the Windows implementation, this track is lightened with respect the original.
Moreover, you do not need to be retro-compatible with previous kernel versions, choose a single version of the kernel and develop on it. If your code works even with older versions it’s better but it is not requested.
You can ask me clarifications about the track, and I can also give suggestions about where to look in the kernel source but you do not need to abuse of my availability and I cannot provide you code snippets.
If it is the first time for you to use git please be careful! You can lose all of your files, always keep a backup and do not use git commands without really knowing what they do!
What you need to hand in
You need to produce the following:
- the kernel implementation of the project, that can be a kernel module or you can just modify the kernel source (both for any kernel version that you desire from 2.4 to 5.12). In any case you need to push the source on your private repo of GitHub classroom;
- a C library with a proper header .h which allows to use your implemented features;
- an example code which shows that your implementation works with more than one scheduler thread and with UMS schedulers which shares worker threads;
- the documentation of the code. You are encouraged to use an automation tool for generating the documentation, like Doxygen or similar and put the output in
- a short final report describing the project decisions and some benchmark of your solution, as the average time needed for a scheduler thread to switch a worker thread or any kind of parameter you want to measure, follow the template in
You do not need to print anything (even if one day we will do exams in presence) and everything must be put in your GitHub Classroom repository, in particular the sources in
/src and the documentation in
What will be evaluated
The evaluation criteria will take into account the following characteristics of your project (in order of importance):
- adherence to the project track regarding the implemented features;
- correctness of the solution;
- questions about the project (asked individually) during the project presentation (you obviously need to know the theory behind the facilities that you use);
- elegance, modularity and cleanliness of the code (you are encouraged to follow how the kernel source is written, e.g. use macros for optimizing things and not functions);
- clarity and quality of the documentation;
- final report and results;
- good usage of git and versioning.
What to do now
For carrying out the project, you need to:
- obviously be joined to the course Google Classroom;
- create a GitHub account, adding your
@studenti.uniroma1.itemail to the list of your addresses;
- accept this invitation link;
- if requested please associate your name with one in the class;
- now create the Team:
- if you will work alone create a team with your name and surname, well written, like “Name Surname”, there is a space between name and surname;
- if you will work in group (max 2) create a team with the two surnames, like “Surname1 Surname2”, there is a space between the 2 surnames.
I trust in your common sense for not adding people that you do not know. Moreover, please be 100% sure of your decision before creating the team;
- start to reason about the project but keep in mind that several topics that will be useful will be presented later in the course
Please plan the discussion date of the project, and set it as your deadline because you need to book the proper exam call on Infostud. See the Exam rules in the course page.
Some suggestion to do the project:
- reason and design the solution before writing code
- use a good kernel source explorer
- use a virtual machine with your custom kernel version and do not perform updates
- use Visual Studio Code
- use the “Test Assignment” of GitHub classroom for playing with the repository, link here
Pull requests for feedback are enabled for your project repository, they are described well here. In practice you can create a pull request for asking clarifications to me but you cannot abuse of this feature and my answers are obviously limited, I cannot provide you code snippets and I cannot do the project for you. The answers to general questions will be made available to everyone in the project track page as F.A.Qs.
- Windows UMS, is the source of the full specification from which this track has been inspired
Can I implement the solution only in User Mode?
No, you cannot. Implementing a solution that tries to solve the problem only in user mode (e.g. you use the module only for /proc) is not enough for passing the practical exam, by doing this you are actually missing the purpose of the project, even if it can implemented.
v4: Added #1 F.A.Q.
v3: Updated information about how to present the project
v2: Solved issue on the description of
/procfiles and folders
v1: First release