- Types of failures:
- Intermittent failures:
- The disk is OK
- Due to electrical fluctuations, a disk read or disk write operation fails
- The disk is OK
- Write failures:
- A Power failure occurs
during a
disk write operation
Result:
- Data in main memory is
lost (due to
power failure
- Data on disk is corrupted
- Data in main memory is
lost (due to
power failure
- A Power failure occurs
during a
disk write operation
- Media decay:
- A sector is
worn
The sector is part of a block and it can no longer be used
- A sector is
worn
- Disk crash:
- The disk head
has scratched the
platter(s)
- Data on the whole disk is lost
- The disk head
has scratched the
platter(s)
- Intermittent failures:
- Technique:
- Verify the data for correctness
- Implementation:
- Every sector contains
a check sum:
Data of 1 sector: +---------------------------------------------+--------+ | data stored in 1 sector | chksum | +---------------------------------------------+--------+
- Read:
success = false; while ( not success ) { read sector; compute checksum of the data; if ( computed checksum == recorded checksum ) success = true; }
- Write:
compute checksum on data; Write data Write checksum Optional: Read the data back and verify....(The optional part is usually not done because of performance hit !!!)
- Every sector contains
a check sum:
- Spare blocks/sectors:
- A disk keep a number of blocks/sectors in reserve (= ununsed)
- Replacing bad sectors/blocks:
- Worn sector/block is marked
as a bad sector/bliock
(Disk maintains a bad sector/block table)
- Bad blocks are not used to store data
Effect of bad sectors/blocks:
- Disk capacity is reduced....
- Worn sector/block is marked
as a bad sector/bliock
- Recall:
- Power failure during a
write operation will:
- Wipe out the copy
of data in
main memory
- May corrupt the copy of data on disk
- Wipe out the copy
of data in
main memory
- Result:
- complete loss of
data
- No way to
get back the old copy of the
data
(unless you have a backup copy)
- Now way to get get back the new copy of the data
- No way to
get back the old copy of the
data
(unless you have a backup copy)
- complete loss of
data
- Power failure during a
write operation will:
- The stable storage update policy:
- For each (logical) sector,
we have two physical sectors:
- Write operation to sector x:
successL = false; ntries = 0; while ( not successL ) { ntries++; write data to sector xL; read data back; if ( verified OK ) successL = true; else { if ( ntries > too-many-times ) { replace xL with another sector; } } } // Repeat the same procedure for xR
- Read operation to sector x:
ntries = 0; while ( ntries < too-many-tries ) { read sector xL; if ( verified OK ) return xL; read sector xR; if ( verified OK ) return xR; }
- For each (logical) sector,
we have two physical sectors:
- Claim:
- Stable storage write policy will preserve the old copy or the new copy of the data during power failure
- Only way to recover
from a disk crash:
- Redundancy (e.g., backup copy)
- Techniques that deploy
redundancy:
- disk mirroring
- RAID (Redundant Array of Independent Disks)
- Mirroring:
- A disk is
a logical unit
- Physically a
disk consists of
two (physical) disks
- A disk is
a logical unit
- Write data to the disk:
- Write data to both disks
- Read data from the disk:
- Read data from any one of disks
- Advantage:
- Read operation is
twice as fast
(Write operation is about the same speed)
Disadvantage:
- Very expesnive (twice the hardware cost)
- Read operation is
twice as fast
- Disk mirroring =
RAID level
1
- RAID level 4:
- Use n disks as follows:
- n − 1 disks are normal disks (i.e., you write data blocks to these disks)
- Disk n is
a parity disk:
- Every sector on disk n stores the result of the parity value of all the (corresponding) sectors on the other n − 1 disks
Example:
- Use n disks as follows:
- Read from a (data) disk:
- Just read the data
from the disk
(No special operation required)
- Just read the data
from the disk
- Write to a (data) disk:
- Write the data to the
disk
- Compute the
new value of
the parity
- Write the new paraity value to the parity disk
- Write the data to the
disk
- Example: write operation
- Read the
data from disk 2:
- Suppose we update the data
to 1111:
- Update the
data disk:
- Update the
parity disk:
- Read the
data from disk 2:
- Problem with
RAID level 4:
- Each write operation will
update the
parity disk
- The parity disk will likely be a bottleneck !!!
- Each write operation will
update the
parity disk
- RAID Level 5 protocol:
- Parity cylinder !!!