Dr. T. Doom 8.1 CEG 433/633 - Operating Systems I Chapter 8: Memory Management A program may not be executed until it is; –associated with a process –brought

CEG 433/633 - Operating Systems I Dr. T. Doom 8.1

Chapter 8: Memory Management

• A program may not be executed until it is;

– associated with a process

– brought into memory

• In allow multi-programming, the OS must be able to allocate memory to each process

– Several processes at once

– Requires a “Memory Management” scheme and appropriate hardware support

– Security?

• The memory management scheme has a large impact upon how a program for a particular platform must be designed and compiled

– How much memory is available?

– How do should we bind addresses?


Address Binding

• Instruction and data addresses in program source code are symbolic:

– goto errjmp;

– X = A + B;

• These symbolic addresses must be bound to addresses in physical memory before the code can be executed

• Address binding: a mapping from one address space to another

• The address binding can take place at compile time, load time, or execution time.

• Compile-time Binding: the compiler generates absolute code

– memory location must be known a priori

– must recompile to move code

– MS-DOS .COM format programs


Load-time Binding

• Most modern compilers generate relocatable object code

– symbolic address are bound to a relocatable address

– i.e. “286 bytes from the beginning for the module doomC.o

• The linkage editor (linker) combines the multiple modules into a relocatable executable

• The load module (loader) is places the program in memory

– The loader performs the final binding of relocatable addresses to absolute addresses

• Load-time Binding: Bind relocatable code to address on load

– Must generate relocatable code

– Memory location need not be known at compile time

– If starting address must change, we must “reload” code


Execution-time Binding

• A logical (or virtual) address space may be bound to a separate physical address space

– Provides an abstraction of physical memory

– Logical (virtual) address – generated by the CPU

– Physical address – address seen by the memory unit

• The user program deals with logical addresses; it never sees the “real” physical addresses

• Memory-Management Unit (MMU): Hardware device that translates CPU-generated logical addresses into physical memory addresses

• Execution-time Binding: Binding delayed until run time

– process can be moved during its execution from one memory segment to another

– logical and physical addresses differ (requires mapping)

– requires hardware and OS support for address mapping


Memory-Management Unit (MMU)

• Logical and physical addresses are the same in compile-time and load-time address-binding schemes; logical (virtual) and physical addresses differ in execution-time address-binding scheme.

• The user program deals with logical addresses; it never sees the real physical addresses.

• Hardware device that maps virtual to physical address.

• In most basic MMU scheme, all logical addresses begin at 0, and the base register is replaced by a relocation register

– The value in the relocation register is added to every logical address generated by a user process at the time it is sent to memory to generate the necessary physical address

– To move the program, simply change the value in the register– The limit register remains unchanged

• Thus, each logical address is bound to a physical address– Is security maintained?


Can we reduce memory requirements?

• Loading: Placing the program in memory

• Dynamic Loading: Routine is not loaded until it is called

– Program must check and load before calling

– If a needed routine is not available in memory, the relocatable linker/loader loads the routine and updates the program’s address tables

• Better memory-space utilization; unused routine is never loaded

– Size of executable is unchanged

– Runtime footprint is smaller

– Useful when large amounts of code are needed to handle infrequently occurring cases.

• No special support from the operating system is required

– Implemented through program design


Can we reduce executable size?

• Linking: combining object modules into an executable

• Most OSes require static linking– All library routines become part of the executable

• Modern OSes often allow dynamic linking– Linking postponed until execution time

• Instead of placing the code for each library routine in the executable, include only a stub (a small piece of code) which:

locates the appropriate memory-resident library routine replaces itself with the address of the routine, and executes

the routine

• Executable footprint is reduced– program will not run w/o libraries– New (minor) versions of the library do not require recompilation

• Some operating systems provide support for sharing the memory associated with library modules between processes (shared libs.)

– Very efficient! No read() required, less overall memory usage


What if there isn’t enough memory?

• How can we execute an executable whose code footprint is larger than the memory available?

– This was a major problem in the 60s and 70s for general purpose computers and remains a major problem

Consider memory usage in an e-mail pager or ISDN box

• Solution: Keep in memory only those instructions and data that are needed at any given time; overload during run-time

– Overwrite this memory with a new set of instructions and data when we get to a significantly different part of the code

– Each set of instructions/data is an overlay

– Programming design of overlay structure is non-trivial

• No special support needed from operating system

– Implemented by user design

• Modern general purpose OSes use virtual memory to deal with this problem


How does the OS allocate memory?

• Contiguous Allocation Scheme: All memory granted to a process must be contiguous

• Single-partition contiguous allocation– Only one “partition” exists in memory for user processes

Only one user process is granted memory at a time The resident operating system must also be held in memory OS size changes as “transient” code is loaded

– Place OS in low memory, use relocation-register to define the beginning of the user partition

Relocation-register protects the OS code and data Alows relocation of user code if OS requirements change

– Relocation register contains value of smallest physical address; limit register contains range of logical addresses – each logical address must be less than the limit register

– To change context, must swap out main memory to a backing store


Swapping

• A process can be suspended and swapped temporarily out of memory to a backing store, and then brought back into memory for continued execution

• Backing store – usually a fast disk large enough to accommodate copies of all memory images for all users; must provide direct access to these memory images

– swap may be from memory (conventional) to memory (extended)

• Roll out, roll in – swapping variant used for priority-based scheduling algorithms (or round-robin with a huge quantum); lower-priority process is swapped out so higher-priority process can be loaded and executed.

• Major part of swap time is transfer time; total transfer time is directly proportional to the amount of memory swapped.

• Requires execution-time binding if process can be restored to a different memory space then it occupied previously

• OS management of I/O buffers required to swap a process awaiting I/O

• Modified versions of swapping are found on many systems, i.e., UNIX and Microsoft Windows


Swapping in Single Partition Scheme


Contiguous Allocation (Cont.)

• For multi-processing systems it is far more efficient to allow several user processes to allocate memory

– The OS must keep track of the size and owner of each partition

– The OS must determine how and where to allocate new requests

• Multiple-partition contiguous allocation

– Fixed-partition: Memory is pre-partitioned, the OS must assign each process to the best free partition

Hard limit to the number of processes in memory Efficient? OS

100K

500K

200K


Contiguous Allocation (Cont.)

• Multiple-partition contiguous allocation

– Dynamic allocation: Memory is partitioned by the OS “on the fly” Operating system maintains information about:

a) allocated partitions b) free partitions (hole) Hole: block of available memory; holes of various size are

scattered throughout memory. When a process arrives, it is allocated memory from a hole

large enough to accommodate it

OS

process 5

process 8

process 2

OS

process 5

process 2

OS

process 5

process 2

OS

process 5

process 9

process 2

process 9

process 10


Dynamic Storage-Allocation Problem

• How do we satisfy a request of size n from a list of free holes. Optimization metrics include speed and storage utilization.

– First-fit: Allocate the first hole that is big enough. Search begins at top of list. Fast search.

– Next-fit: Allocate the first hole that is big enough. Search begins at the end of the last search. Fast search.

– Best-fit: Allocate the smallest hole that is big enough; must search entire list, unless ordered by size. Produces the smallest leftover hole.

– Worst-fit: Allocate the largest hole; must also search entire list, unless ordered by size. Produces the largest leftover hole.

• Simulation shows that:

– First-fit is better (in terms of storage utilization) than worst-fit

– First-fit is as good (in terms of storage utilization) than best-fit

– First-fit is faster than best-fit

– Next-fit is generally better than first-fit


Fragmentation

• How do we measure storage utilization? – How much space is wasted?

• Internal fragmentation – allocated memory may be slightly larger than requested memory; this size difference is memory internal to a partition, but not being used

– Problem in fixed-partition allocation

• External fragmentation – total memory space exists to satisfy a request, but it is not contiguous.

– Problem in dynamic allocation– 50% rule: Simulations show that for n-blocks, n/2-blocks of

memory are wasted. 1/3 of memory is lost to fragmentation– External fragmentation can be reduced by compaction

Shuffle memory contents to place all free memory together in one large block

Compaction is possible only if relocation is dynamic, and is done at execution time and if the OS provides I/O buffers so that devices don’t DMA reallocated memory


Non-Contiguous Memory Allocation

• Goal: Reduce memory loss to external fragmentation without incurring the overhead of compaction

• Solution: Abandon the requirement that allocation memory be contiguous.

• Non-contiguous memory allocation approaches include: – Paging: Allow logical address space of a process to be

noncontiguous in physical memory. This complicates the binding (MMU) but allows the process to be allocated physical memory wherever it is available.

– Segmentation: Allow the segmentation of a process into many logically connected components. Each begins at its own (local) virtual address 0.

This allows many other useful features, including protection permisions on a per segment basis, etc.

Example segmentation: Text, Data, Stack.– Segmentation with Paging: Hybrid approach


Paging

• Physical memory is broken up into fixed-size partitions called frames

• Logical memory is broken up into frame-size partitions called pages

• The OS keeps track of all free frames

– Frame size = Page size (power of 2, usually 512 - 8k bytes)

– To run a program of size n pages, need to find n free frames and load program

– Internal fragmentation (average of 50% of one page per process)

• Logical addresses must be mapped to physical addresses

– Set up a page table to note which frame holds each page

• Logical Address generated by CPU is divided into:

– Page number (p) – used as an index into a page table which contains base address of each page in physical memory

– Page offset (d) – combined with base address to define the physical memory address that is sent to the memory unit


Paging Example


Implementing Paging

• Paging is transparent to the process (still viewed as contiguous)

• Divide a m-bit logical address for a system with pages of size 2n into:

– n-bit page offset (d)

– (m-n)-bit page number (p)

page number p page offset d

m - n bits n bits

m-bit logical address

• The page number p is an index to the page table which stores the location of the frame

• Frames and pages are the same size, thus the displacement within a page is also the displacement within the frame

• Mapping is:– Physical address = page-table(p) + d


Address Translation Architecture


Page Size

• How large should a page be?– Smaller pages reduce internal fragmentation– Larger pages reduce the number of page table entries

• If s is the average process size, p is the page size (in bytes) and e is the # of bytes per page table entry, then:

s/p: # pages / processse/p: size of page table / processp/2: memory lost to int. fragmentation

Overhead = se/p + p/2

Mimimize: dp(overhead) = -se/p2 + 1/2 = 0 or p = sqrt(2se)

• For current process sizes, and available physical memory, optimal page sizes range between 512 - 8K bytes

• Page table must be kept in main memory.– Why? If a page is 8k (12 bits) and the CPU uses a 32-bits

address then there are 220 possible pages per process– # of bits per entry depends upon size of physical memory– The memory consumed by this table is overhead/waste


Implementation of Page Table

• The page table must be kept in main memory– Page-table base register (PTBR) points to the page table

add PTBR + page number (p) to get lookup address– Page-table length register (PRLR) indicates size of the table

Only make the page table as large as necessary Addresses in unallocated pages cause an exception

• For each CPU memory access in there are two physical accesses– access the page table (in memory) to retrieve frame– access the data/instruction

• The inefficiency of this two memory access solution can be reduced by the use of a special fast-lookup hardware cache for the page table

– associative registers or translation look-aside buffers (TLBs)

• Hit Ratio: The percentage for which the necessary data is present in the cache

– otherwise, get data from page table in main memory


Effective Access Time

• Effective Access Time (EAT) is a weighted average

tTLB: time required for a TLB lookuptmem: time required for an access to main memory: hit ratio

EAT = ( tTLB + tmem) + (1- )(tTLB+tmem+tmem)

• Even for fairly small TLBs, hit ratios of .98 - .99 are common

– Most programs refer to memory very sequentially and locally

– The 32-entry TLB in the 486 generally has a .98 hit ratio

• Thus, we can implement paging without suffering a significant latency cost

• Try it with TLB search of 20ns, Memory access of 100ns, and hit ratios of .80 and .98


Memory Protection

• Protections bits are included for each entry in the page table:– Valid-invalid bit indicates if the associated page is in the process’ logical

address space, and is thus a legal page Machines which have a PTLR can avoid the “wasted” page table

entries necessary to house the i bit.– RO/RW/X bits indicates if the page should be considered read-only,

read-write and/or executable– Protection exceptions are calculated in parallel with the physical

address (after the page table lookup)

• Page tables allow processes to share memory by having their page tables point to the same frame

– Note: Processes can not reference physical memory that the OS does not allow them to via page table setup

• The OS keeps a frame-table (one entry per frame) which indicates if each frame is full or empty, to which process the frame is allocated, when was it last referenced, etc

– Memory protection implemented by associating protection bit with each frame


Shared Pages

• Private code and data – Each process keeps a separate copy of the code and data

• Shared code– To be sharable, code must be reentrant (or “pure”)

All non-self modifying code is pure - it never changes during execution (I.e. read only code)

Each process has its own copy of registers and data storage to hold the data for its process’ execution

– One copy of reentrant code can be shared among processes (i.e., text editors, compilers, window systems)

– Problem: Shared code must appear in at the same location in the logical address space of each process

internal branch and memory addresses must be consistent


Shared Pages Example


Two-Level Paging

• Consider a page table for a 32-bit logical address space on a machine with a 32-bit physical address space and size 4K pages

– logical space/page size = 232 / 212 = 220 entries

– physical space/frame size = 232/212 = 220, 20 bits/entry + ~12 protection bits ~= 4 Bytes/entry

– Page table size = 220 entries * 4 Bytes/entry = 4 MB

– 4 MB >> 4K: The page table itself is larger than one page!

– We can’t allocate the page table in contiguous memory

• We must page the page table! The page number is divided into:

– How many 4 Byte entries per 4K page? 212/22 = 210

a 10-bit page offset

– How many bits remain? 20 - 10 = 10 a 10-bit page number

• Thus, a logical address is divided pi, an index into the outer page table, and p2, the displacement within the page of the outer page table

page number page offset

pi p2 d

10 10 12


Two-Level Page-Table Scheme


Multilevel Paging Performance

• The concept can be extended to any number of page-table levels

• Since each level is stored as a separate table in memory, covering a logical address to a physical one may take many memory accesses

• Even though time needed for one memory access is increased, caching (via TLB) permits performance to remain reasonable

• Example: In a system with a two-level paging scheme, a memory access time of 100ns, and 20ns TLB with a hit rate of 98 percent:

effective access time = 0.98 x (20 + 100)

+ 0.02 x (20 + 100 + 100 + 100)

= 124 nanoseconds.

which is only a 24 percent slowdown in memory access time.


Inverted Page Table

• Problem: Each process requires its own page table, which consists many entries (possibly millions). How can we reduce this overhead?

• Solution: The number of frames is fixed (and shared between the processes). Store the process/page information by frame!

– One entry for each “real” page of memory

– Entry consists of the virtual address of the page stored in that real memory location, with information about the process that owns that page

• Concern: Decreases memory needed to store each page table, but increases time needed to search the table when a page reference occurs

– Use hash table to limit the search to one — or at most a few — page-table entries

– hash table requires another memory lookup (of course)

• Concern for later: The use of an inverted page table does not obviate the need for a normal page table in demand paged systems (ch. 9)


Inverted Page Table Architecture


Segmentation

• Segmentation is a non-contiguous memory allocation scheme

– “simpler” than paging, but not as efficient

– supports user view of memory

• Programmers tend not to consider memory as a linear array of bytes, they prefer to view memory as a collection of variable sized segments

– Never forget, however, that memory is a linear array of bytes

• A segment is a logical unit such as:

– main program, procedure, function, local variables, global variables, common block, stack, symbol table, arrays, etc.

• Segmentation is a memory management scheme that supports this user view of memory

– segments are numbered and referred to by that number

– a logical address consists of a segment, and an offset

– A mapping between segments and physical addresses must be performed


Logical View of Segmentation

1

3

2

4

1

4

2

3

user space physical memory space


Segmentation Architecture

• Logical address consists of a two tuple:

<segment-number, offset>,

• Segment table – maps two-dimensional physical addresses; each table entry has:

– base – contains the starting physical address where the segments reside in memory.

– limit – specifies the length of the segment.

• Segment-table base register (STBR) points to the segment table’s location in memory.

• Segment-table length register (STLR) indicates number of segments used by a program

segment number s is legal if s < STLR.


Segmentation Architecture (Cont.)

• Relocation– dynamic (execution-time)– by segment table

• Sharing– similar to sharing in a paged system

shared segments must have same segment number in each program

– protection/sharing bits in each segment table entry

• Memory allocation – segment vary in length– dynamic-storage problem: first fit/best fit?– external fragmentation

segmentation don’t use frames, thus external fragmentation exists

periodic compaction may be necessary and is possible as dynamic relocation is supported


Sharing of segments


Hybrid: Segmentation with Paging

• Segmentation and paging have their advantages and disadvantages

– segmentation suffers from dynamic allocation problems lengthy search time for a memory hole external fragmentation can waste significant resources

– paging reduces dynamic allocation problems quick search (just find enough empty frames if they exist) eliminates external fragmentation

– Note: it does introduce internal fragmentation

• Solution: page the segments!

– First seen in MULTICS, dominates current allocation schemes

• Solution differs from pure segmentation in that the segment-table entry contains not the base address of the segment, but rather the base address of a page table for the segment segment offset

s p d’

18 6 10

page


MULTICS Address Translation Scheme

segment offset

s p d’

18 6 10

page


Generalized Summary

• Parkinson’s Law: “Programs expand to fill available memory”

• Mono-programmed systems:

– One user process in memory

– OS and device drivers also present

– Overlays used to increase program size

– Relocatable at compile-time only

– Protection: Base and limit register

• Multi-programmed systems/fixed number of tasks (OS/360 MFT):

– Memory allocation on fixed-sized/numbered partitions

– Queue for each partition size

– Relocatable at load time

– Protection: Base and limit register, or protection code (pid) if multiple non-contiguous blocks are allowed


Generalized Summary

• Multi-programmed and time-shared systems with variable partitions

– Memory manager must keep track of partitions and holes

– Dynamic allocation algorithm: First-fit, Next-fit, Best-fit, etc.

– Compaction to reduce external fragmentation

– Protection: relocation (base) register and limit register, or virtual addresses - the OS produces the physical address;

user programs can not generate addresses which belong to other processes

– Relocatable during execution (or no compaction possible) Change relocation register value or page-to-frame mapping

Documents

Dr. T. Doom 8.1 CEG 433/633 - Operating Systems I Chapter 8: Memory Management A program may not be executed until it is; –associated with a process –brought