[mythtv-users] Hard Drive reliability esp. RAID issues

[John Drescher]

On Mon, Mar 8, 2010 at 9:25 PM, Harry McGregor <hmcgregor at biggeeks.org> wrote:
> -----BEGIN PGP SIGNED MESSAGE-----
> Hash: SHA1
>
> On 3/8/10 7:15 PM, John Drescher wrote:
>>> For my home use, I use RAID6, with 14 1TB drives, plus a hot
>>> spare.
>>>
>>> I am considering moving towards some raid10 groups instead for
>>> greater performance.
>>
>> Do you need better performance? A 14 drive linux software array of
>> 1TB drives should net you at minimum 500MB/s writes with 600 to 800
>> MB/s reads.
> That is in sequential read/write
>
> When you start throwing a lot of streams at it, think 3-6 tuners, 3
> front ends, maybe a few com flag processes, you can still have IO
> stall conditions.  If front end X wants Y blocks, and has to wait too
> long for them, you will see skips on the the player.
>

I see. That is probably caused by the combination of forced flushes
and database activity neither of which make good use of the cache.

Take a look at /sys/block/mdX/md/stripe_cache_active and if it is
being maxed out during writes increse stripe_cache_size.

John