Hi Brian,
You wrote:
> Ignoring CPU, filesystem, and library limitations, what is the theoretical
> maximum size of a raid device?
>
> The raid superblock appears to store the block size of each physical
device
> in an unsigned 32-bit integer allowing a maximum physical device size of
> 4 terabytes. If the stock raid device code is used, limiting the maximum
> number of physical devices per raid device to 12, and we configure all
twelve
> drives to be used in a raid-0 array, is it true that in theory the raid
device
> is 48 terabytes large (52,776,558,133,248 bytes).
The md driver manipulates buffer_head structures which use an unsigned long
to refer to the device's real sector. Since most systems use 32 bit longs
and 512 byte sectors, the md limit cannot be any bigger. 2^(32+9) = 2^41 =
2,199,023,255,552 bytes.
> I would like to believe that my calculations are correct but I'm highly
> skeptical; seeing how the multidisk source only uses 64-bit integers for
the
> event counter. This leads me to believe that the maximum size of a raid
> device is limited to the 32-bit integer used to store the physical device
> size...hence 4 terabytes is the maximum.
The event counter counts superblock updates and its value will be hopefully
very low. I have no idea why it is 64 bits--this causes trouble in the
superblock structure being at a 4 byte (not 8) boundary. A lowly 32 bit
event counter would allow you to hot-add and hot-remove disks every second
for almost a century.
<>< Lance.
