[mythtv-users] .23/.24 NFS mounting / Storage Groups

[Brian Wood]

On Wednesday, October 20, 2010 05:23:41 pm Scott wrote:
> On Oct 20, 2010, at 5:22 PM, Brent Bolin wrote:
> > My current setup is a back end with raid disk storage and tuners.  I
> > also have a back end running in a Xen environment as a guest helping
> > with commercial flagging.
> > 
> > I also just upgraded my network to GB today.
> > 
> > Flagging from the mythtv guest helps take the cpu load off the primary
> > back end, but still puts a stress on disk I/O.
> > 
> > I've been toying with the idea of creating another storage directory
> > for mythtv recordings and that being an NFS mount from the primary
> > back end.  One of my main concerns is creating a complected
> > configuration.
> > 
> > Is this the only way to spread disk usage between systems(NFS)?
> 
> Are the disks really stressed during commflag? I have a single disk SATA
> 7200 RPM drive that records 3 HD streams from a HDHR while doing commflag
> at the same time. Disk IO has never been a bottleneck on the system.
> 
> If you do want to spread the work, your best bet for simplicity is more
> physical drives in the same system. You can use mdtools or LVM to strip
> data across two disks. to spread the work.

Stating the obvious, but:

Linear RAID or RAID 0 will "spread the load" and increase apparent disk performance, but also increase the probability of 
losing the array's contents due to a disk failure, as those systems provide no redundancy.

RAID1 or higher can increase performance (generally read more than write) without additional risk to data, bdue to the 
redundancy those systems provide.

As for how much commflagging stresses disks, it depends on a lot of factors, like how fragmented the files being flagged are, 
the size and bitrates of the files, and other factors.

But probably the easiest way to minimize the possible bottlenecks caused by disk performance is to have the OS and 
database on a separate spindle from the media files, as has been suggested here for a long time. If the system works well 
with everything in one place, than the "if it works leave it alone" rule obviously applies.

Only if this doesn't work should more exotic measures be taken, at least IMHO.