PERFORCE change 123662 for review

[Pawel Jakub Dawidek]

On Fri, Jul 20, 2007 at 03:07:16PM +0200, Alexander Leidinger wrote:
> Quoting Ulf Lilleengen <lulf at FreeBSD.org> (Fri, 20 Jul 2007 14:35:24 +0200):
> 
> [growing RAID-5]
> > Well, what I do is to attach/create the new subdisk as usual, but since it's a
> > RAID-5 array that I know is operational, I give the subdisk a flag, and sets the
> > plex in a resize state. Then, In the raid-5 code, I modify gv_raid5_offset
> > (which basically computes offsets within a subdisk based on the number of
> > subdisks and stripesize). However, what I do, is that instead of taking all
> > subdisks in the calculation, I only take those who does not have the GROW flag
> > (when reading), and I take all subdisks into calculation when it's a write.
> > 
> > This means, that if a create a gv_grow_plex function that reads (stripesize x
> > sdcount) bytes (from the subdisks that do not have the GROW flag), and writes
> > that data to the plex (including all subdisks). This way, i sort of overwrite
> > the old data, but the data is spread out over the new subdisks. I'm sorry if
> > this might seem a bit complex, but just ask more questions if you didn't
> > understand.
> 
> Do you use the additional drive(s) only to write checksums to them, or
> do you write real data to it? If the later, how do you make sure you
> read the right data in case you read data again, which was just written
> there a moment before (how do you know to read from all subdisks and
> not only from a subset in this case)?

You only need to move offset while you synchronize new disk.

When you start you have:

	Disk0	Disk1	Disk2	NewDisk

	D0	D1	P0	U
	D2	P1	D3	U
	P2	D4	D5	U
	D6	D7	P3	U
	D8	P4	D9	U
	P5	D10	D11	U

After some time you have:

	Disk0	Disk1	Disk2	NewDisk

	D0	D1	D2	NP0
	D3	D4	NP1	D5
	U	U	U	U
-->	D6	D7	P3	U
	D8	P4	D9	U
	P5	D10	D11	U

And at the end you have:

	Disk0	Disk1	Disk2	NewDisk

	D0	D1	D2	NP0
	D3	D4	NP1	D5
	D6	NP2	D7	D8
	NP3	D9	D10	D11
	U	U	U	U
	U	U	U	U

Where:
D<x> - data block
P<x> - parity block
NP<x> - new parity block
U - unused
--> - if offset in I/O request is below that point, you use four disks,
      if it is above that point you use only three disks

BTW. Such functionality is really cool.

-- 
Pawel Jakub Dawidek                       http://www.wheel.pl
pjd at FreeBSD.org                           http://www.FreeBSD.org
FreeBSD committer                         Am I Evil? Yes, I Am!
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 187 bytes
Desc: not available
Url : http://lists.freebsd.org/pipermail/p4-projects/attachments/20070723/82baa8cb/attachment.pgp