"Allistar" <> wrote in message
news:8tudneBJj_g5-...
> Lawrence D'Oliveiro wrote:
>
>> As hard drives get larger and larger, more and more RAID
>> configurations
>> become no longer worth using. Nobody should be using RAID 5 any
>> more. RAID
>> 10 may still be worth something, but for how long?
>>
>> <http://blogs.zdnet.com/storage/?p=483>
>>
>> Roll on the next-generation filesystems ...
>
> RAID isn't so much about file systems, it's about either performance
> (by
> parallelising reads) or redundancy (in case a disk dies).
>
> I don't see how a new file system will help if the disk dies.
>
Do we even need a new file system?
Consider this (I've tried to simplify the arrangement so no comments
on whether this would be a USEFUL arrangement please!):
If I have three physical disks (let's assume 750GB each for ease of
calculation) and use a hardware RAID controller to set them up in
RAID5.
I now have what the OS sees as a single 1500GB disk.
If I now have the OS 'split' that into three virtual disks of 500GB
each, and have the OS create a software RAID5 array across those three
virtual disks, then I now have what the user / apps see as a single
1000GB disk.
Now, if one of the physical disks dies and gets replaced, the hardware
RAID would rebuild that from the parity info on the other two physical
disks.
If it runs into an URE and cannot rebuild one of the sectors, that
sector is lost completely. Okay.
However, if we consider the software RAID, this will manifest itself
as a sector on ONE (?) of the three virtual disks having been lost or
damaged (experienced an URE), but this is not fatal since it can
rebuild that sector from the parity information on the other two
virtual disks.
Is that correct, or am I missing something?
Thanks,
Alan.
--
The views expressed are my own, not those of my employer or others.
My unmunged email is:
(valid for 30 days
min probably much longer).