Lecture 11 - More on Scheduling

Last time we talked a bit about scheduling algorithms. We keep the discussion going.

Examples of Preemptive Scheduling

Priority Scheduling: Give each job a priority (rank, price, etc.) and choose the most important one. The idea is to break the priorities into classes:

• IO Bound first (they block)
• Then compute bound (they don't block, faster to compute)

For example, there's CTSS: Compatible Time Sharing System:

• Uses a large quantum for CPU-bound jobs
• Process that use the whole quantum move down
• Process doing IO move up to the top again

But you can grant priority based on:

• Process already in memory? Give higher priority.
• Locks held/needed
• Resource requirements
• Shortest remaining work yet

Another example would be lottery, where you just say "LOL completely random" and be random with the processes.

Minix Scheduling

Minix uses the following structure:

• Layer 1: Written in Assembly. The other layers are in C.
• Layer 2: IO Processes, one per device type. Compiled w/ the kernel
• Layer 3: Server process: File System, Memory Manager, etc. They never terminate while the system is running
• Layer 4: User level

Minix 2 uses the following scheduling schema per level:

Minix 3 added a 16-level priority scheduling system:

• Variable quantum size set in the process table
• A process that uses the entire quantum will be given another one and sent to the end of the queue
• Potential change of priority level for processes using complete quanta
  • If it uses its whole quantum and is its own successor: drops a level
  • If not, rises a level
• Process that block with time remaining in their quantum are moved to the head of the queue when they return, but only with the remaining part of their quanta

Three-Level Scheduling

For any scheduling scheme we schedule at three different locations/types:

• admission: The system can decide whether or not to admit
• memory: Constraints are different for jobs in memory and on disk
• disk: (above)