[mythtv-users] OT Hardware RAID 5 non-Recommendations

Steven Adeff adeffs.mythtv at gmail.com
Thu Dec 21 14:04:16 UTC 2006

On 12/20/06, David Rees <drees76 at gmail.com> wrote:
> On 12/20/06, Marco Nelissen <marcone at xs4all.nl> wrote:
> > >not even close. I could barely get 1 HD 1 PVR-150 and 1 HD playback
> > >with my original RAID5 setup, I since went to a RAID 10 with 4 drives
> > >and its finally capable of handling all that along with my SQL
> > >database. I'm looking to add two more drives, at which point I think I
> > >can go back to RAID5 and still get the speed I need and gain a lot of
> > >extra space (which would be *lovely*.
> >
> > What on earth was your original RAID array made of? Floppies? My single
> > drive mythtv box handles recording multiple HD streams while playing
> > back one just fine. Any modern harddrive should be able to handle that.
> With raid 5 doing a lot of small writes is often very slow unless you
> have a hardware raid controller with a decent amount of onboard
> memory. Raid 5 is slow for small writes because each write requires a
> read to generate parity data so it's not uncommon to see a Raid 5
> array under random small write load to perform worse than a single
> disk.
> Because mythtv writes and flushes data out to disk in small chunks (I
> wish you could configure how mythtv flushed data to disk) it can
> really kill raid 5 performance.
> Raid 10 doesn't suffer from this since you never have to perform any
> reads to perform a write so your typical Raid 10 array will write n/2
> times as fast a single drive and read n times as fast as a single
> drive in the best case scenario (where n is the number of drives in
> the array).
> Raid 5 read performance usually scales up linearly with the number of
> disks (best case), write performance for large writes should scale to
> n-1 (no parity needs to be read from disk).
> Keep in mind that those are best case numbers and other factors
> usually come into play so  real life performance is usually a decent
> amount slower than the theoretical limit.
> For example, there seems to be a software Raid 5 performance bug which
> limits write performance to single disk speeds even for large writes:
> http://thread.gmane.org/gmane.linux.raid/13932/focus=13932

perhaps I'll stick with RAID10 then...

Before you ask, read the FAQ!
then search the Wiki, and this list,
Mailinglist etiquette -

More information about the mythtv-users mailing list