[mythtv-users] RAID 1 for recordings and livetv

[Joseph Fry]

> - If individual disks, the recordings are written sequentially on each
>> disk, with minimal head movement.
>>
> ...
>
>  Just my recommendation... I'm sure others would argue otherwise.
>>
>
> No, seems a reasonable "layman's guide". Some slight
> corrections/observations though :
>
> When writing a recording, the file is synced about once/second. That
> involves a write of the data and then a write of the metadata. So each
> second there will be as a minimum "seek-write-seek-write" per stream. Some
> of the metadata writes will involve multiple seeks - perhaps reading some
> data on free space, updating that (and it's metadata), and then writing the
> data.
> So it's actually worse than the "one seek per thread per second"
> suggested.
>

I over simplified things... but the end result holds true... the number of
seeks on each drive increase if you put the drives in a stripe set rather
than balance the load across independent disks.


> RAID1 can turn in excellent read performances - potentially much higher
> than a single drive.


Absolutely true... however it can never outperform an equal number of
non-raided drives, assuming a equally distributed load, and write
performance will suffer in raid1, lower than the slowest disk in the
array.  There is always some overhead with raid.


> In the absence of writes, an intelligent controller can distribute read
> requests across the two drives according to their head locations - so seek
> distances can be significantly reduced on both drives and it's possible to
> think of a contrived (and artificial) situation where a 2 disk stripe could
> give more than twice the read performance of each disk individually.
>

While I suspect that someone could devise a scenario where you could
accomplish this, I have never heard of an array being faster than the
combined performance of it's parts.  Obviously this is assuming that the
load can be properly spread across the individual disks.  This is where
raid will always win... if I am doing a single file read or write... I
cannot balance this across multiple disks... so raid will be faster.


> Of course, throw in writes and both disks must seek to the same place -
> but then can seek off in different directions to satisfy reads from either
> end of the mirrored volume.
> But I suspect it's going to be rare where the advantages of a striped
> volume outweigh the downsides for Myth use.
>

Indeed, a very well designed raid controller can accomplish amazing things
to optimize the performance of the array... but typically this only applies
to situations where the array is explicitly designed for the intended
load.  For example, I know that several manufacturers make arrays that
perform very well with MS Exchange mail... but MS is now recommending
Exchange be deployed using JBOD distributed across all of the servers
rather than using a SAN... give better performance and fewer points of
failure.


Fun stuff, raid.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://www.mythtv.org/pipermail/mythtv-users/attachments/20121005/1e334496/attachment.html>