Magnetic Disks Have cylinders, sectors platters, tracks, heads - - PowerPoint PPT Presentation

Magnetic Disks

Have cylinders, sectors platters, tracks, heads virtual and real disk blocks (x cylinders, y heads, z sectors per track) Relatively slow, much slower than RAM very mechanical High rate of failure Sizes have gone up A LOT!

RAID

Redundant Array of Inexpensive (or Independent) Disks Improve performance by using multiple disks in parallel. Have several disks seem like one. Disks also have tendency to fail

• more disks, more failures likely
• providing more reliability would be

a good thing too.

RAID

Comes in several different levels (0-6) providing varying degrees of performance and/or redundancy at varying costs. Striping (RAID-0) – no redundancy

• offers high data transfer and I/O

throughput

• suffers lower reliability and availability

than a single disk.

RAID

Mirroring (RAID-1) – uses equal amount

• f disk capacity to store original and

its mirror.

• all writes also go to the mirror
• provides redundancy of data and
• ffers protection against loss in the

event of physical disk failure.

• reads can be done round-robin for

better performance.

• can have multiple mirrors (n-way)

RAID

Can combine RAID0 and RAID1 Commonly done. 0+1 or RAID10 example: stripe six disks and have six more for a mirror. What happens when a disk goes bad in a mirror and has to be replaced? How likely is data loss?

RAID

RAID-2 uses bitwise striping across disks and used additional disks to hold Hamming code check bits. Can correct single-bit errors Can detect double-bit errors Used in CM-2, not much else

RAID

RAID-3 uses a parity disk to provide redundancy.

• Stripes data across all but one disk in
• array. Uses other disk to store parity
• info. (XOR)
• Can recover from a single data disk

failure.

• How is that possible? How to figure
• ut data stored on failed disk?

RAID

RAID-4 attempts to provide higher rate

• f data transfer by spreading I/O load

as evenly as possible across all disks in the array. Maps data and uses parity the same as RAID3 by striping the data across all disks and XORing the data for the info

• n the parity disk.

The difference between RAID3 and 4 is that 3 access all the disks at one time and 4 access each disk independently.

RAID

The RAID4 way allows the array to execute multiple I/O requests simultaneously while RAID3 can only execute one I/O request at a time. RAID4 performs reads much better than writes. The parity disk can become a bottleneck for writes as all writes update it. Need to fix parity disk bottleneck.

RAID

RAID-5 is similar to RAID-4 except the parity is spread throughout the

• disks. This does away with the parity

disk bottleneck. Need n + 1 disks

D0 D2 P2 D1 P1 D4 P0 D3 D5

RAID

RAID-6 is similar to RAID-5 except two parity checks are done for more reliability. Can withstand two disk failures without losing data. Need n + 2 disks Software RAID and Hardware RAID

Disk scheduling

Disks are slow with mechanical movements involved. 3 factors to time

• seek time – moving arm to right

cylinder/track

• rotational delay
• data transfer time

seek time usually dominates

• try to reduce average time
• try to reduce the head movements

Disk scheduling

Assume a queue of disk block requests

• n different cylinders.
• try to optimize seek time
• minimize cylinders traversed

First-Come First-Served (FCFS) Serve requests in order they come in Shortest Seek First (SSF) – handle closest request next.

Cylinders at edges could suffer starvation

Disk scheduling

SCAN or elevator algorithm

• also used in buildings with elevators
• go in one direction and handle each

request you come across

• turn around and go in the other

direction

• alternatively always go in one direction

and go back to 0 after reaching end

• C-SCAN – Circular SCAN

I/O devices

In Unix look like files

• can be read, written, etc with sys calls
• /dev, /devices
• block and character special files
• major and minor device numbers

Berkeley sockets for networking TCP, UDP