Emery Berger

1
Operating SystemsCMPSCI 377Lecture 16 File
Systems II

• Emery Berger
• University of Massachusetts, Amherst

2
Outline

• File Systems Implementation
• How disks work
• How to organize data (files) on disks
• Data structures
• Placement of files on disk

3
Disk Surface Layout

• Tracks concentric rings on disk
• bits laid out serially on tracks
• Tracks split into sectors or blocks
• Minimum unit of transfer from disk

4
Disk Pack Multiple Disks

• Disks organized in disk pack stack of platters
• Use both sides of platters
• Two read-write heads at end of each arm
• Cylinders matching sectors on each surface

5
Cost of Disk Operations

• In addition to CPU time to start disk operation
• Latency time to initiate disk transfer
• Seek time time to position head over correct
  cylinder
• Rotational time time for correct sector to
  rotate under disk head
• Bandwidth rate of I/O transfer of sectors once
  initiated

6
Outline

• File Systems Implementation
• How disks work
• How to organize data (files) on disks
• Data structures
• Placement of files on disk

7
File Organization on Disk

• File system maps file blocks to disk location
• (file 0, block 0) (platter 0, cylinder 0,
  sector 0)
• Key performance issues
• Support sequential and random access
• What data structure is best for maintaining file
  location information?
• How do we lay out files on the disk?

8
On-Disk Data Structures

• File descriptor structure used to describe where
  file is on disk and attributes
• Must be stored on disks just like files
• Most systems fit following profile
• Most files are small
• Most disk space taken up by large files
• I/O operations target both small large
• Per-file cost must be low, but large files must
  also have good performance

9
Contiguous Allocation

• Operating system maintains ordered list of free
  disk blocks
• OS allocates contiguous chunk of free blocks when
  it creates a file
• Only need to store start location size in file
  descriptor

10
Contiguous AllocationPros Cons

• Advantages
• Simple
• Access time? Number of seeks?
• Disadvantages
• Changing file sizes
• Fragmentation? Disk management?
• Examples IBM OS/360, write-only disks, early PCs

11
Linked Files

• Maintain list of all free sectors/blocks
• In file descriptor, keep pointer to first
  sector/block
• In each sector, keep pointer to next sector

12
Linked Files Pros Cons

• Advantages
• Fragmentation?
• File size changes?
• Efficiently supports which type of access?
• Disadvantages
• Does not support which type of access? Why?
• Number of seeks?
• Examples MS-DOS

13
Indexed Files

• OS keeps array of block pointers for each file
• User or OS declares maximum length of file
  created
• OS allocates array to hold pointers to all blocks
  when it creates file
• But allocates blocks only on demand
• OS fills pointers as it allocates blocks

14
Indexed Files Pros Cons

• Advantages
• Wastes very little space
• Sequential random access easy
• Disadvantages
• Sets maximum file size
• Lots of seeks (why?)

15
Multilevel Indexed Files

• Each file descriptor contains 14 block pointers
• First 12 pointers point to data blocks
• 13th pointer one indirection
• Points to block of 1024 pointers to 1024 more
  data blocks
• 14th pointer two indirections
• Points to block of pointers to indirect blocks
• Used in BSD UNIX 4.3

16
Multilevel Indexed FilesPros Cons

• Advantages
• Simple to implement
• Supports incremental file growth
• Small files?
• Disadvantages
• Indirect access inefficient for random access to
  very large files
• Lots of seeks (data not contiguous)
• Is file size bounded?
• What could OS do to get more contiguous memory?

17
Free-Space Management Bitmaps

• Need free-space list to keep track of which disk
  blocks are free
• Akin to free-list for main memory
• One approach bitmap
• One bit per block on disk
• Bit 1 , block is free
• Easy to find any page free in next 32 bits
• Marking block free very simple

18
Free-Space Management

• Problems
• Space Bitmap has big RAM footprint
• 80GB disk, 512 byte sectors 5Mb
• Performance Slow if most of disk is in use
• Alternative link together free blocks
• Head of list cached in kernel memory
• Each block contains next pointer
• Cost of allocation? Freeing? Allocating
  consecutive blocks?

19
Summary

• Many of concerns implementations of file
  systems similar to virtual memory implementations
• Contiguous allocation simple but...
• External fragmentation, compaction, relocation...
• Indexed allocation page tables
• Free space managed with bitmap or linked lists

20
Next Time

• I/O Systems