Disk Management Flashcards
Disk Access Time
For b bytes:
Ta = Ts + 1/(2r) + b/(rN)
- N = bytes per track
- Ts = seek time
- r = spinning speed
- b = bytes to access
Seek Times: 4 phases
- speed-up
- coast
- slowdown
- settle
Skewing
Goal: keep read speed high
=> shift sectors on consecutive tracks somewhat -> reduce rotational delay
Sparing
Map faulty sectors to other sectors via a list passed to the controller
Disk Scheduling Algorithm: FIFO
= Process requests in order of arrival
-> fair
-> very efficient for clustered operations
-> inefficient for random reads
Disk Scheduling Algorithm: Shortest Seek Time (SSTF)
= process first request with smallest seek time
-> requires estimation of seek time
-> high utilization/efficiency
Disk Scheduling Algorithm: SCAN
= SSTF with arm changing direction when reaching the extremes of the disk
-> less starvation than SSTF
-> lower delay variance, higher average delay
-> unfairness center vs. extreme tracks
Disk Scheduling Algorithm: C-SCAN
= SCAN but once extreme is reached, continue from the other extreme
-> reduces unfairness
Disk Scheduling Algorithm: LOOK
= SCAN but the arm changes direction if there are no more requests in the current direction
-> slightly more efficient than SCAN
-> more complex
-> unfairness center vs. extreme tracks
Disk Scheduling Algorithm: C-LOOK
= C-SCAN but changes direction if there are no more requests in the current direction
Disk Scheduling Algorithm: V-SCAN
Process first request which is closest located with:
distance = SSTF distance + 1 (if changing direction) * R * width of the disk
- VSCAN(0) = SSTF
- VSCAN(1) = LOOK
- VSCAN(0.2) gives good balance between SSTF and LOOK
Disk Scheduling Algorithm: FSCAN
Use a 2-stage buffer to handle I/O
Apply SCAN on the requests in the first buffer, then process requests in the second buffer
RAID
= Redundant Arrays of Independent Disks
=> many inexpensive disks form one large, fast and reliable disk
RAID 0
Data equally striped over N disks:
Disk i contains blocks i, i+N, i+2N,..
-> poor robustness
effect of stripe length:
- large: random reads simultaneous
- small: sequential read fast
RAID 1
Data mirroring over 2 discs
-> all data on both disks
Very high robustness but expensive solution
Profit on random reads but write is slower than with one disk