CPU is faster than main memory, we need to add a ”gearbox” between the cpu and the memory; cache!
Base + limit registers protect processes from each other. (Used for comparison at addressing)
Giving a user program a physical adress space
(Using a programming analoug here…)
Compile time; absolute code is generated (”constant” addresses, recompilation is needed if changed).
Load time; relocatable code (”static” addresses, can change, but not without reloading user code)
Execution time; (”variable” addresses; can change at runtime)
Logical = cpu’s view
Physical = memory unit’s view
Address space; the set of all addresses as seen by a user (logical) or the corresponding cpu (physical)
Mapping between theses is done by MMU
Relocation register (in the MMU)
Dynamic loading (load only when used); programmers responsibility
Dynamic linking; at execution time using a ”stub”
Swap out a process to free memory (and resources).
Swapped out process; in a ready queue
Swap time (one way!) o (process_size / backing_store_transfer_rate) + latency + seek_time
Time quantum > swap time; if not, time is wasted.
Only swap idle processes!
Contigous allocation
Partitions; fixed/variable size
Allocation algorithms o First-Fit, Best-Fit,Worst-Fit
External/Internal fragmentation (possible solution; compaction)
If non-contiguous allocation is ok;
Logical adress: page number (translated into a frame number via page table) + offset = physical address
Page/frame-size = hardware defined o Logical address space = 2 m bits, page size 2 n
bits, give a logical address: page number = m-n and page offset = n
Small page size = less internal fragmentation; if too small = I/O overhead
size of page table entry: 4 Bytes (32 bits) -> 2
32
physical page frames
Registers for page table
Big tables; store table in main memory (pointed to by the Page-Table Base
Register) and introduce TLB to reduce overhead in memory access. o Effective memory access time =
(TLB_hit_ratio * mem_access) + (TLB_miss_ratio*(mem_access[page table] + mem_access[data]))
Protection; associate extra bits to each frame-entry
Problem : large page tables (2
32
logical address with page size 4 Kb
2
32
/2
12
entries
(1 milion) in the page table. 4 Byte entries = 4 MB PER PROCESS!
Solution : page the page table!
Split the page number bits in two (page + page[off_page_table] offset)
So, 64 bit architectures? With 3 (!) level paging we may have a page size of
2
34 bits. Do’h! o Introduce hashing
Inverted page table
Problem
: addresses are like a long array of bytes. User’s view; different segments, some contain function, other variables (and so on).
Solution : something very similar to basic memory management
Split a logical address space into segments.
Let the user use a two-dim. address; <segment-number; offset>
Segment table
Registers; segment base and limit
Segmentation fault (sounds familiar, eh?)