Re[2]: Concept problem with RAID1?

Re[2]: Concept problem with RAID1?

[Jim Klimov]

Hello PFC,

Friday, March 24, 2006, 7:07:04 PM, you wrote:

P>      I think you would like something like this :

P>      A LVM (or dm- device mapper) layer which sits between the RAID layer and
P> the physical disks. This layer computes checksums as data is written to
P> the physical disks, and checks read data against these checksums.

P>      Problem is, where do you store the checksums ?
Why not store the checksums on the same device that stores the
blocks? For example, let a 64-kb (65536b) block of data store,
say, 65520 bytes of data and 16 bytes of checksum/digest and a
timestamp. If the hardware goes crazy and some bits (or more)
are mixed up, we can detect if the data is not trustworthy -
the checksum and/or data is screwed and they don't match each
other. A drawback is the impossibility to mount such a device
as-is, and some loss of performance in addressing/buffering
which I think is okay as a tradeoff for reliability.

A good thing (arguably) would be a capability of raid1 layer
to interact with this checksummed layer during rebuilds or even
casual work. If a block is found to be bad on one submirror,
it could be read (and re-written) from another, valid, copy
without remaking the whole metadevice.

-- 
Best regards,
 Jim Klimov                            mailto:klimov@2ka.mipt.ru

Re: Re[2]: Concept problem with RAID1?

[John Rowe]

A much nicer way to get that sort of reliability would be for RAID6 to
periodically scan the blocks on the device and to use the extra
information to do ECC (and for RAID5 to at tell syslog).

John