Next: Distributed Systems Up: 4560 Previous: Page Replacement Algorithms Contents

File Management [204]

files are logical units mapped onto physical secondary storage
file name: logical object
physical objects: blocks on disk, tape, optical disk
one or more sectors: smallest unit to read from or write to disk
block: unit of I/O transfer from disk to memory
- improves efficiency
secondary storage: nonvolatile
file attributes: type, location, size, protection, time, date, user ID
volatility: frequency of additions and deletions
activity: percentage of records accessed during time frame
directories: keep track of files (and are files themselves)
create the illusion of compartments
but are indexes to files which may be scattered
entry per file: name, attributes, disk address, etc.

Files: Abstract Data Type [205]

data (the file) and the operations on the file
create: allocate storage, add to directory
open: search directory
- establish logical pointer current-file-position
write: data at the pointer
read: data at the pointer
reposition: the pointer (seek)
delete: file entry in directory
truncate: update length attribute in directory
append: new data and update length attribute in directory
rename: change name in directory
close: disconnect logical access to file
access methods:
sequential: only increment pointer
direct: set pointer for seek
indexed: index of keys and associated direct pointers

Files: Layered Systems [206]

application programs

logical file system: symbolic file name (directories)

file-organization module: logical block addr TO physical block addr

basic file system: read/write physical blocks

I/O control (device drivers): memory TO/FROM disk system

devices

Files: Partitions - Directories on Disk [207]

$\begin{picture}(565,305)(10,495) \thicklines\put(435,495){\framebox (105,305){}}... ...225,600){\makebox(0,0)[lb]{\raisebox{0pt}[0pt][0pt]{partition B}}} \end{picture}$

partitions = minidisks = volumes
virtual disks

Files: General Graph Directory [208]

$\begin{picture}(744,295)(20,535) \thicklines\put(260,785){\framebox (180,35){}} ... ...\put(240,590){\makebox(0,0)[lb]{\raisebox{0pt}[0pt][0pt]{cyclic}}} \end{picture}$

single-level: one directory shared by all users
two-level: one directory per user
tree-structured: long pathnames
graph-structured: sharing

Files: Contiguous Allocation of Space [209]

simplest method is a contiguous set of blocks
disk address of start (base) block and length (in blocks)
sequential access is easy - read next block
direct access (``seek'') is easy -
logical offset = physical base + offset
how to find space for new file?
first-fit: first hole that is big enough
best-fit: smallest hole that is big enough
external fragmentation
blocks are available but not sequentially
internal fragmentation
preallocation of blocks is too large
left-over amount in last block

Files: Contiguous Allocation [210]

$\begin{picture}(530,480)(40,340) \thicklines\put( 80,780){\framebox (20,20){}} \... ...t(425,795){\makebox(0,0)[lb]{\raisebox{0pt}[0pt][0pt]{directory}}} \end{picture}$

Files: Linked Allocation [211]

solves external fragmentation: can use any block for any file
solves ``preallocation'' internal fragmentation
good for sequential access: chase the pointer
not effective for direct access
where is the nth block of the file?
requires space for pointers
if a pointer is bad, file is lost

Files: Linked Allocation [212]

$\begin{picture}(520,480)(40,340) \thicklines\put( 80,780){\framebox (20,20){}} \... ...} \put(380,660){\makebox(0,0)[lb]{\raisebox{0pt}[0pt][0pt]{mail}}} \end{picture}$

Files: Indexed Allocation [213]

contiguous: easy sequential and direct access but fragmentation
linked: no fragmentation but difficult direct access
indexed: direct access and no fragmentation
bring all pointers into one location: the index block (IB)
each file has its own IB
ith entry in the IB points to the ith block
suffers from wasted space (IB may not be full)

Files: Indexed Allocation [214]

$\begin{picture}(560,480)(40,340) \thicklines\put( 80,780){\framebox (20,20){}} \... ...t(475,780){\makebox(0,0)[lb]{\raisebox{0pt}[0pt][0pt]{directory}}} \end{picture}$

Files: How Large an IB? [215]

each file must have an IB - IB should be small
some files are small - IB should be small
some files are big - IB should be large
solution: multiple levels of smaller IBs
some files are small - don't need multiple levels
solution: UNIX Index Block
1 direct block of pointers to data (good for small files)
indirect blocks (i-nodes) for multiple levels

Files: UNIX inode [216]

$\begin{picture}(600,460)(30,360) \thicklines\put(240,800){\framebox (50,20){}} \... ...0,615){\makebox(0,0)[lb]{\raisebox{0pt}[0pt][0pt]{direct blocks}}} \end{picture}$

Files: Free-Space Management [217]

free-space list records all blocks that are free
allocate blocks (create, append): search list and remove
deallocate blocks (delete): add to list
implementations:
bit vector: bit=1 IMPLIES block is free
grouping
first block has n-1 addresses of free blocks and 1 address of the next group
counting: blocks are often given up in contiguous sets
keep a list of pairs (first block, count)
linked list: add/delete blocks at head

Files: Linked Free-Space List [218]

$\begin{picture}(320,480)(40,340) \thicklines\put( 80,780){\framebox (20,20){}} \... ...\makebox(0,0)[lb]{\raisebox{0pt}[0pt][0pt]{free-space list head}}} \end{picture}$

Files: Protection - Access Matrix [219]

domain / object	F1	F2	F3	printer
D1	read		read
D2				print
D3		read	execute
D4	read/write		read/write

process in domain D1 can read file F1
matrix is sparse
access (column) list - each object has list of domain, rights-set
process in domain D_i accesses O_j: scan j's list for permission
capability (row) list - each domain has list of object, rights-set
process in domain D_i possesses a capability to allow access
list is a protected object and is accessed indirectly

Next: Distributed Systems Up: 4560 Previous: Page Replacement Algorithms Contents

Ted Billard 2001-11-17