Assignment on modifying XV6 kernel

Pulak Malhotra, 2019101050 Note: The original commits can be found in Assignment5 Repository in this repo.

Implementation:

1. Changing proc structure and time command

• Added ctime, etime, iotime and rtime
• Changed ctime in allocproc and etime in exit
• Updated rtime in timer interrupt of trap function
• Added waitx system call which gets the rtime and wtime of first zombie
• Implemented time command using exec and waitx system call

2. PS command

• Made a system call called get_ps() which prints all the detail of the proc structures with pid > 0 after acquiring the ptable lock
• Implemented ps command which just calls this system call
• Further improvement : dont print directly from kernel

3. Schedulers

• I have a different file for each scheduler that gets compiled according to the given flag value. Each scheduler has its own scheduler and trap function
• FCFS
  • Do not call yielding in trap
  • Use ctime to get min ctime process and run that
• PBS
  • A set prio sys call to change the priority
  • Run the process with the least priority
  • after running the process find the min prio (if it changed during execution)
  • run the process with min prio again
• MLFQ
  • The proc structure now stores time of entry (toe) of each process, cur_q, time spent in each queue, ticks it has left before getting preempted
  • Go through all process with decreasing priority to find one in runnable state and run it
  • After it is run and it completed the time slice move it down a queue
  • If it did not complete time slice dont change its queue
  • Also check for aging in all processes
  • Aging wait time is currently (1 << (level + 4)) for a level
  • In trap only yield if the process has done its alloted ticks

I ran the benchmark.c file for all the schedulers and observed the following in it: Benchmark spawns total 50 processes which alternate between the following types:

• Pure CPU
• Pure IO
• First CPU then IO
• First IO then CPU For each type of scheduler I have noted down the end time of the different types of processes and the total time for benchmark process.

MLFQ

• Time 3194 ticks
• First pure IO
• Then IO then CPU
• Then CPU then IO
• Then pure CPU

FCFS

• Time 3841 ticks
• First CPU bound (in ascending order)
• Then mixed CPU then IO and IO then CPU (in ascending order)
• Lastly Pure IO (in ascending order)
• There is some interleaving among categories but for each type they follow ascending order strictly

PBS

• Total time 3255 ticks
• First all odd procceses since they have lower priority by definition in benchmark
• Then all even processes
• Among odd and even
  • First Pure IO
  • Then CPU then IO
  • Then IO then CPU
  • Then pure CPU

RR (default)

• Total time is 3224 ticks
• First Pure IO
• CPU then IO
• IO then CPU
• Pure CPU

BONUS