dwswager said:
This sounds stupid and is kinda like trying to figure out if the light
in the fridge goes out when you close the door.
P5AD2-E Premium
SIL3114 SATA RAID controler
2-200 GB Seagate 7200.7 SATA Drives
I connected everything up. The SIL ROM Utility recognized the drives
and I selected a mirror and had the utility automatically setup the
array.
When I went to the Disk Managment in windows XP it ran the new Disk
wizard. Windows only shows 1 disk as I would expect if the controller
is handling the mirror functionality. I setup a new volume and it only
presented the options of simple volume and not RAID Mirror. Windows
shows the disk as Dynamic, 186.31GB NTFS, Healthy.
Question? How does one tell that both drives are getting the data such
that there is backup if one disk fails?
I think this is an excellent question. Someone posted a problem
to this group once, where a disk in their mirror failed, and
the data on the remaining drive was missing data/out of date.
I can only presume that the status shown in Windows all that t
ime was "healthy", when in fact it was not.
The mirror state is based on a premise. The premise is that,
exactly the same operation is done to each disk. If the RAID
hardware and software detect a diverge of this assumption, then
the status of the array should immediately change, to reflect
that the assumption is no longer true.
This means, when the motherboard POSTs, both disks must answer
queries in the first 30 seconds or so. If a disk fails to
respond, the RAID BIOS/software must mark the remaining disk
as an orphan, and tell the user the array is broken.
Similarly, if the mirror is doing a write operation, and one
disk fails to complete the write operation, that should result
in the array being broken.
Corner cases can occur, if the computer is not powered by a
UPS. If the power is cut to the computer, when one disk has
succeeded in writing its sector, and the second disk has not,
there is no way to record the divergence of the two disks.
What happens then, might very well depend on whether the
file system is journalled or not. In any case, the disks could
diverge from that point onwards.
A similar situation could occur, if the RAID chip you are using
is a "soft RAID". For example, I have read that the SIL0680 is
a soft RAID. The thing is, if the RAID driver crashes, and has
only written data to one disk, that is the same thing as the
power failed case. I would at least want a solution where the
"mirroring" is done purely by hardware.
If a RAID chip is truly "hardware", the RAID chip will make
sure that both disks get written, or it will present a status
that there is a failure. If the software driver to the hardware
chip fails, if will fail "atomically", as either a single command
is given to the hardware chip to write both disks, or the
single command has not been given.
There is, I suppose, the possibility that a bus crash (arbiter
failure) could freeze out the RAID hardware chip, before it can
complete identical operations on both drives.
In summary, I think you can see there are many possibilities
for the array to become desynchronized. Perhaps taking the
array offline once in a while, and checksumming all the files
on the disk would help. For this to work, you would have to
find an IDE or SATA interface, that is not going to be
confused by the presence of the "reserved sector" on the disk.
If you are using the Promise 20378, for example, you could
flip the BIOS setting from "RAID" to "IDE", connect one
drive, and create a manifest for the disk. (A manifest is
a directory listing, plus a checksum for each file.) Repeat
for the other disk. Run a chkdsk (NOT repair option) on
each disk, to see if any files are damaged from an abrupt
termination before a file was completely written, or a
directory structure was updated.
Based on the results of the test, you could choose to return
to RAID land, reenable RAID operation, break the array, and
re-clone one disk to the other. Because you have done chkdsk,
a surface scan, a manifest, you would be reasonably sure that
the cloning operation will succeed.
Which is a hell of a lot of work, and explains why I for one
am not interested in a mirror, when I can have a backup image
taken once in a while instead. If mirroring software was capable
of doing a consistency check on both volumes of a mirror,
perhaps my opinion would change. As long as the only automated
option is to re-clone one disk to the other, you'll never
know whether hidden faults are collecting on one of the disks.
And tearing the array down, to checksum it, is not going to be
a "free' operation, if the user has to do it.
Paul
