the function of parity disk in RAID system

parity disk is used in RAID 3 & 4, how it is used in RAID system?

for example:

Reading: no r/w operations on parity disk.
Writing: whenever writing a block to disk, the corresponding block on
parity disk should be updated.
Fault Recoving: using blocks on parity disk to calculate the lost
data.


above is my own understanding, I don't know if it's correct, can you
give me some instruction? thank you.

Laser wrote:
parity disk is used in RAID 3 & 4, how it is used in RAID system?

for example:

Reading: no r/w operations on parity disk.

Correct.

Writing: whenever writing a block to disk, the corresponding block on
parity disk should be updated.

Correct.

Fault Recoving: using blocks on parity disk to calculate the lost
data.

Correct. The recovery algorithm takes the surving blocks plus the parity
block and XORs them together to recreate the missing information.


--
Nik Simpson

But how does it update the parity information when writing into one
block on data disk? I guess there are 2 approaches exists:

1. read out corresponding blocks on all other data disks, exclusive-or
with the new data block, then write to correponding block on parity
disk.

2. read out the old data before writing, and read out corresponding
block on parity disk, then exclusive-or old data with new data and
parity block, and write back to parity disk.

which of above approches was implemented into actual use. why choose
it?

(Laser) writes:
But how does it update the parity information when writing into one
block on data disk? I guess there are 2 approaches exists:

1. read out corresponding blocks on all other data disks, exclusive-or
with the new data block, then write to correponding block on parity
disk.

Right. This is recomputing the parity from scratch.

2. read out the old data before writing, and read out corresponding
block on parity disk, then exclusive-or old data with new data and
parity block, and write back to parity disk.

Right. This is computing the parity difference between old & new data.

which of above approches was implemented into actual use. why choose
it?

Both! The tradeoff depends on how much of the stripe (across all data
disks) is being written. If less than 50% of the data on the stripe
is being replaced, it's cheaper to read the old data being replaced
and do the parity-difference computation. If more than 50% is being
replaced, it's cheaper to read the non-replaced data and do the
parity-from-scratch computation.

In practice, some controllers always choose one strategy or the other,
but most can do either. The decision point isn't always set at
exactly 50%, either.

Have you read the original RAID papers? Worth a look. I believe this
is described in them.

D. A. Patterson, G. Gibson, and R. H. Katz. A Case for Redundant
Array of Inexpensive Disks (RAID). In Proceedings of ACM SIGMOD,
pages 109-116, Chicago, IL, June 1988.

-- Anton