[mythtv-users] An LVM'd drive died! What do I do...

Blammo blammo.doh at gmail.com
Sat Nov 12 12:35:39 EST 2005


On 10/27/05, Brian C. Huffman <huffman at graze.net> wrote:
> On Thu, 2005-10-27 at 11:42 -0400, Bryan Halter wrote:
>
> > Well to start with for RAID you need to have disks that are all the same
> > size (preferably the same model).  I believe Linux supports growing
> > software RAID volumes and I'm sure someone will correct me if it
> > doesn't.  Personally I'd go out and buy a 4 device SATA-RAID controller
> > and 4 250GB drives that will give you 750GB of storage and fewer
> > headaches since the RAID array will be seen as any other scsi disk and
> > the card will do the thinking so you won't take a CPU hit for having
> > RAID.


I just wanted to throw in my 2cents into this 360' discussion about
raid and file systems.

Nothing is perfect.

For data integrity and speed, HW Raid > SW Raid > no raid


First I had no raid, just a single 300g drive. I found that with 4
tuners going, and 3 machines trying to commerical flag, watching even
a single frontend was impossible without either stuttering video,
choppy FFW, "IO Bound" on disk writes, network saturation, or some
combo of the above. So I went to SW Raid. Ran XFS (tried EXT3, JFS,
XFS, ReiserFS)  / NFS (tried 2,3,4)  / SWRaid 5.

 All the stuttering, IO Bound issues went away, and I could get good
recordings, but stability just wasn't there. I fought with that for
almost 2 years. I never lost "all" the data, which is obviously the
goal of RAID5. I had IDE controllers die, individual drives die,
motherboards die, upgraded multiple versions of Kernel, OS, etc, and
always fought little crash issues, kernel panics, etc. I'd put the
overall stability about 90%, which is TERRIBLE for something with a
24/7 duty cycle.  I tried VIA chipset MB's, NV chipset MB's, MB IDE,
MB SATA, PCI SATA, PCI IDE, and any combo. Just before I made the jump
to HW Raid, I was up to 4xIDE 4xSATA, running on an Abit N7G2.

Then about 3 months ago, when I got fed up with taking calls at work
from the SO like "Honey, the backend locked up again", I finally
bought a HW Raid card. (3ware 9500-s12)

WOW.

I should have done that in the beginning.

Gone, INSTANTLY,  all the stability issues. period. CPU load drops to
10% under heavy IO load.

It took me almost two days to migrate all the data off to a couple of
300G HD's build a new RAID5 array and copy it all back, while
continuing to do daily recordings. But wow.

Yes, there are plenty of things still that could make my life
miserable. I dodged a bullet just yesterday. Laying in bed, I hear a
"THUMP'. Come to the computer room and the backend is off. I smell
something in the air..... Long story short, the Antec Truepower 550
gave up. Slapped a new Enermax 600W in, and we're back up and
operating, no data loss. Yes, I realize the PS could have coastered
all my drives and I would have lost everything, raid or not.

However, for me, this solution gives me exactly what I need. Speed,
Stability, and the ability to sustain a SINGLE drive failure without
loss of total data. (which seems to happen about every 6 months....
consumer drives are NOT rated for 100% duty cycle, but that's another
discussion)

Looking at the graphs from my switch (snmp & jffnms are a great combo)
with commerical flaggers going, recorders going, I see sustained
periods of 300Mbit  on the main gig port of my primary backend. All
without a hitch. Stable, solid, and days of uptime.

So, my advice, in the short? Buy good hardware, set it up in ways
known to be stable, and it will serve you well. You get what you pay
for.

Hope this helps someone else going through this journey.


More information about the mythtv-users mailing list