Mass Storage Systems - Southern Oregon University

Download Report

Transcript Mass Storage Systems - Southern Oregon University

Moving-head Disk Mechanism
Rotation Speeds: 60 to 200 rotations per second
Head Crash: read-write head makes contact with the surface
Disk Structure
•
•
•
•
•
Cylinder: the set of tracks that all the heads are currently located at.
Track: A ring on a disk where data can be written
Sector: The smallest transfer unit of data accessed in a block
Cluster: A group of sectors the operating system treats as a unit
Organization Choices
– Sector mapping (One dimension array of logical blocks)
• 0 is first sector, track 0 of the outermost cylinder.
• Subsequent sectors map through tracks, through cylinders, in an outer
to inner direction.
– Sector counts and density
• fixed sectors per track: varying densities
• Varied sectors per track: outer tracks have more sectors; constant
density
– Bad block management
• Sector sparing: replace bad sectors with spares in the same cylinder
• Sector slipping: copy all sectors down to the next spare
Magnetic Tape
•
•
•
•
•
•
Early secondary-storage medium of choice
Persistent, inexpensive, and has large data capacity
Very slow access due to sequential nature
Used for backup and for storing infrequently-used data
Kept on spools
Transfer rates comparable to disk if read write head is
positioned to the data
• 20-200GB are typical storage capacities
Storage Attached to a Host System
• Basic: Connections are accessed through I/O ports
communicating through busses
• Small Computer System Interface (SCSI) protocol
–
–
–
–
Connects to sixteen devices
One SCI initiator and up to fifteen targets
Each target controls up to 8 logical units
Logical units can be RAID arrays or a CD jukebox
• Fibre Channel (FC) is high-speed architecture
– Private high speed switched network to many attached hosts
– Storage can be dynamically allocated to each hosts
• Other interface Standards: USB, EIDE, ATA
Network-Attached Storage (NAS)
• Storage made available over a network
• Protocols
– NFS is a common protocols
– iSCSI embeds SCSI protocol in IP.
• Implementation: remote procedure calls (RPCs)
Storage Area Network
• Popular for large storage environments
• Private high speed network
• Flexible for attaching many hosts to multiple storage arrays
Basic Storage System Concepts
• The OS presents a one of two virtual storage abstraction
– Raw device: an array of data blocks (a partition).
– File system: OS schedules interleaved application requests
• Goal: optimize disk access time and throughput
• Access time (seek time + rotational latency)
– Seek time: moving the heads to the cylinder with the data
– Rotational latency: rotate the disk head to the desired sector
– Transfer rate: data flow speed between drive and computer
– Operating systems attempts to minimize seek time; hardware
optimizes rotational latency
• Definitions
– Throughput: bytes transferred per unit time
– Disk bandwidth: total bytes transferred / (time from first
request to time of last transfer)
First Come First Serve (FCFS)
Service the first request first
This Illustration shows a total head movement of 640 cylinders.
Shortest Seek Time First (SSTF)
Service the request closest to the current head position
• SSTF scheduling is similar to SJF scheduling
• Starvation of some requests is possible
• This Illustration shows total head movement of 236 cylinders.
SCAN (Elevator) and C-Scan
SCAN: Service the first request encountered in the current head direction
C-Scan: Services in one direction; SCAN: Services in both directions
C-Scan: Return to the beginning after each pass (circular list)
Head going left
Movement = 236 cylinders
Head going right
Movement = 382 cylinders
• C-Scan: Provides a more uniform wait time than SCAN.
• Repositioning to the beginning is faster than repositioning in
small pieces because of acceleration and deceleration.
Look and C-Look
Similar to SCAN and C-SCAN except don’t go to end of disk on each pass
Head going right
Movement = 322 cylinders
C-Look Algorithm
Which Disk Scheduling Algorithm?
• SSTF applies if starvation is not an issue (light
loads)
• SCAN and C-SCAN perform better with
heavy loads
• Requests for disk service can be influenced by
the file-allocation method
• The disk-scheduling algorithm
– Normally a separate background running module of
the operating system
– It can easily be replaced if desired
Disk Management
• Factory Low-level formatting
– Divide disk into sectors for
access by a disk controller
• OS Partitioning
– Divide disks into separate
groups of ‘raw’ cylinders
– Sector sparing is a
technique to recover from
“bad” blocks
• User Logical formatting
– Initialize the file system
MS DOS Structure
Disk Booting in Windows 2000
The ROM bootstrap reads the boot sector,
which then proceeds to load the OS
Swap-Space Management
• Swap-space: A disk extension of main memory
• Implementation can be:
– a normal part of the file system (slower but flexible)
– in a separate disk partition (faster but less flexible)
• Examples
– 4.3 BSD
• Allocates swap space when process starts
• holds text segment (the code) and data segment
– Solaris 2: allocates swap space only when a page is forced
to disk
Swap Space
Virtual memory extension of main memory
BSD Program (Text) Segments
BSD Data Segments
Redundant Array of Inexpensive
Disks (RAID)
• Goal: achieve reliability through redundancy or increase speed
through parallelism
• Seven Raid Organizations
0: Non-redundant striping: Parts of data stored on different disks. Group of disks
acts as a single storage device
1: Mirrored data: stores (shadows) duplicates of each disk
2: Error correction codes: Parity is spread over a group of dedicated disks
3: Bit interleaved parity: Parity is written to a dedicated disk
4: Block interleaved parity: Parity stored on a disk separate from the data
5: Block-interleaved distributed parity: stripes data and parity across the RAID
6: P + Q redundancy: Parity and data are written, both with redundancy
Parity: Information to reconstruct data in case of failure
RAID
Examples
• Top: striped and
then mirrored
• Bottom:
mirrored and
then striped
Stable Storage
Definition: Storage that carries a “guarantee” of reliability
• Transactions and write-ahead logs require stable storage
• To implement stable storage:
– Replicate information on more than one nonvolatile storage
media with independent failure modes.
– Update information in a controlled manner to ensure that we
can recover the stable data after any failure that happens
during data transfer or recovery.
– After failure, copy incomplete copies from the media that has
successfully completed its operation.
Tertiary (Removable) Storage
• Characteristic: Low cost, high capacity
• Examples
– Removable magnetic disks: nearly as fast as hard disks, but less reliable
– A magneto-optic disk: records on rigid platter coated with magnetic material
• The read write heads are further from the head, resistant to head crash
• lasers used to amplify magnetic material for write and detect
– Optical disks: Altered by laser light. Very durable and reliable.
– Tape: Falling price gap between tapes and disks, makes these less popular
• Hierarchical Storage Management (HSM)
–
–
–
–
–
Extends primary and secondary storage to include tertiary storage
Found in supercomputing centers that have enormous volumes of data
Usually implemented as a jukebox of removable disks
Disk-resident files are archived to a tape or disk library for low cost storage.
Files are automatically staged back (slowly) to disk for active use
Note: fixed drives tend to be more reliable than removable ones
Cost
Price/Megabyte Hard Disk
From 1981 to 2000
Price/MB DRAM
From 1981 to 2000
Price/Mb Tape, From 1984-2000