CS 3733 Operating Systems

Lecture 12 [2/9/98]: Introduction to Memory Management

Reading: SG 8.1 - 8.4

Address binding

Fundamentals of memory management on a single user system

Program is compiled into machine code. Uses a relocatable format for most addresses and there is a table at the beginning for those addresses which are not relocatable.
When the program is loaded into memory all of the appropriate relocable addresses are adjusted.
If dynamic loading is used, the entire routine is kept in relocatable format and loaded in when called. This is done by the program.
Dynamic linking --- used for system libraries. Also called shared libraries. The program contains only a stub. The stub loads the appropriate library when called. This is a feature of the operating system.

Multiprogramming

Keep more than one process in memory.
CPU utilization = 1 - pⁿ where n is the number of processes and p is fraction of time each process spends in the I/O wait state.

Fixed partitions for memory management

Divide up the memory into fixed pieces (not necessarily equal).
Each partition contains at most one program.
A base register and a limit register are used to relocate and protect.
When job is scheduled it is put into a particular partition.
If job is a lot shorter than partition, that space is wasted for duration of job (internal fragmentation).
If a partition is too small for any of the current jobs, that partition sits idle (external fragmentation).
If job is larger than any partitions, you're out of luck.

Variable partitions

Partition sizes are determined dynamically to match the current job sizes.
The operating system keeps a table of current parititions.
When a job finishes, it leaves a hole. Eventually there are holes everywhere which must be consolidated (external fragmentation). This is called compaction.
Hardware is same as for fixed partitions (base register and a limit register).

Swapping

Move a process out of memory to disk and then bring it back into memory to continue its execution.
There is too much overhead to have the CPU scheduler constantly swap out processes. (How long would it take to swap out a 500 K process to a disk with a transfer rate of 3 Mbytes/sec?) Problems:
- The CPU scheduler must work on a shorter time scale than this.
- Even if done in the background (by a separate I/O controller) this would produce too much disk activity.
- Some processes, like X Windows applications are over a megabyte in size.
- We cannot swap out a process which is not using the CPU because it is doing I/O.
Swapping can be used by the long term scheduler to swap out processes that are waiting for an event that will not occur for a long time.

Revision Date: 2/6/98