[linux-audio-user] AMD64 question: update

[Pieter Palmers]

Paul Winkler wrote:
> On Mon, Jul 03, 2006 at 08:07:45PM -0500, Jack O'Quin wrote:
>> We all have anecdotes about good and bad disks we have known.
>> My impression is that all the manufacturers have trouble with various
>> new technology at various times.  I've heard good and bad stories
>> about them all from time to time through the years.
> 
> Yep. 
> I've had cheapo drives work without problem until I upgraded and
> threw them out. I've had a carefully researched drive purchase
> completely die without warning after 2 years of light use. 
> 
> Since then, I try to assume that any drive by any manufacturer
> can die at any time.  MAKE BACKUPS.

It should also be noted that a lot of hard disk failures are due to 
mains power loss.

Modern harddisks use a lot of write caching on the controller to achieve 
decent performance. So when power goes down when there is data in the 
write cache, it is lost. The file system however 'thinks' that data has 
been written correctly. This hence results in file system corruption.

But there is also the issue of interruption of the actual physical write 
operation when power goes down. If a drive is writing a sector when the 
power goes down, it won't be able to update the CRC data for the sector, 
hence the next time the sector is read, the CRC check fails and the 
drive marks the sector as 'bad'. The sector is however not mechanically 
broken. When doing a low level format of the drive that also tries to 
re-initialize the bad sectors, the drive will be repaired to an 'as new' 
state.

Another important note that this is *not* solved with RAID mirroring. In 
fact, RAID mirroring will make things worse, as you now have two 
independent write caches and head locations.


One solution to this is to disable write caching on the drive, and live 
with the consequences, mainly a very big performance hit.

My personal solution is a UPS. Most of the power outages for me are due 
to some circuit overload, so the 15 minutes of battery power are enough 
to survive the re-enabling of the circuit. And if not it shuts down 
cleanly.

My experience is that *every time* your system has a power failure, you 
have some file system and hard disk corruption.  I would certainly 
recommend everybody to use one, a UPS is a better investment than a 
second disk for software RAID mirroring. And it is pretty cheap (about 
100€ here).

An unrelated remark with respect to hard disk configuration:
I personally think that using a hardware RAID5 controller (e.g. a 3ware 
escalade), together with a bunch of less-performing but more 
silent/reliable drives will give you the best solution.
I think something like a RAID-5 of 4x 100Gb 5400RPM disks will give you 
a system that is fast, reliable, and high capacity (300Gb). A shame that 
5400RPM drives are so hard to get...

I've tried a lot of setups and a lot of harddisks, all resulting in 
failures due to power outages. Since I switched over to a 3ware escalade 
RAID5 setup (3x160GB maxtor) and a UPS, I haven't had any problems. I 
can only recommend it. Of course it is not the quietest of setups, but 
it's not that bad... my laptop is noisier.

Greets,

Pieter