User-Space-OOM

Compile and Run the project with following steps:

0. git checkout master
1. Build the kernel module in MemoryKernelModule directory using the make command. The output file memory_info.ko will be generated.
2. Build the PSI Interface file with the make command in PSI_Monitor file. The output file PSI_Monitor will be generated in that directory.
3. Build the user-space OOM daemon with the make command in UserSpaceOOM directory. The output file userOOM will be generated in that directory.
4. Build the test program in test_proc1 directory with make command. The output file userOOMtest will be generated in that directory. The utility of this process is 8.
5. Build the test program in test_proc2 directory with make command. The output file userOOMtest will be generated in that directory. The utility of this process is 5. We now have two test processes with assigned priorities that we will run to check which one gets killed earlier. Note that - Process 2 is more memory intensive than Process 1.
6. In order to run the ML Processes, following steps need to be executed in the command line.

 pip3 install pytorch
 sudo dnf install opencv 
 pip install opencv-python

7. Insert the memory_info kernel module using command sudo insmod memory_info.ko. dmesg will show the status of the module. On sucessful insertion of the module, you will see the percentages of available memory and swap memory printed in the kernel log(dmesg).
8. Create new folder user_processes in your /tmp directory. We will be dynamically storing the process priorities in this directory.

cd /tmp
mkdir user_processes

9. Now run the userOOM daemon [cmd = ./userOOM]. This will keep on running in that terminal window until you manually terminate by ^Z.
10. Now run the PSI Interface in another window PSI_Monitor [cmd = ./PSI_Mem]. The set threshold for fullavg10, poll interval and recovery interval will be displayed along with continuous monitoring of full avg10 values. The can be manually terminated by ^Z.
11. In a new test window, now run the bash script testproc.sh to run two ML Test processes and dummy memory allocaton processes respectively.

bash testproc.sh

The two heavy ML processes will be exacting on the memory along with the two dummy test programs that will keep allocating memory until system runs out of memory. The user-space daemon will kill the processes such that the utility as well as the memory pressure released is maximized. All test programs prints their PIDs which can used to see if userOOM kills the right task. The statistics of the running process such as OOM_Score, Value and VmRSS will be displayed. The command line will display if a process is killed. The test processes that were killed will be displayed along with the reduction in overall system OOM Score and the associated expense of killing these processes measured by their utility values.

In the current run, the knapsack thresholds are set such that one ML process and the two dummy programs will be terminated