Re: RAW devices

DA Morgan wrote:
> Joel Garry wrote:
> >> Just received my new NetApp 270C yesterday: Dual filer heads and a
> >> TB of disk. It took less than 4 hours to build a 4 node cluster on it.
> >
> > Software RAID-4? How long does it take to rebuild parity when you lose
> > the parity drive and another drive? How do you know how long the
> > batteries will last in nvram when the batteries in your ups die?
> >
> > (One starts thinking about these things when one works on systems more
> > than 2 years old and has to deal with the consequences of no one
> > watching the batteries and arrays of inexpensive disks manufactured in
> > one batch reaching their time of failure, which is much less than the
> > rated mtbf. But that is too late, of course).
> >
> > jg
> > --
> > @home.com is bogus.
> > It is now a crime to delete files.
> > http://catless.ncl.ac.uk/Risks/24.20.html#subj6
>
> A production system should not rely on batteries. It should rely
> on dual diesel generators with fuel purchased from two independent
> companies. And the rest of what this implies.

So what does it rely on before the diesels fire up? And how do you know the generators will always work right? I've seen both fail spectacularly (and worse, not-so-spectacularly so the problem is not noticed right away).

http://groups.google.com/group/comp.databases.oracle/msg/29f1684073028648?oe=UTF-8&output=gplain http://groups.google.com/group/comp.databases.oracle.server/msg/0aea2237897d488a?dmode=source&hl=en

>
> I've never lost a parity disk in production where we didn't have
> it mirrored on a second system. You might want to direct your inquiry
> to NetApp.

Well, I"ve seen it. That wouldn't be NetApp specific (I have no experience with them), and I wouldn't expect them to admit anything technically, since I doubt Mr. Murphy works _for_ them. ;-)

http://toasters.mathworks.com/toasters/2411.html

The issue with both disk drives and memory is you just can't know how abused they consider themselves over time, even if you have dedicated environments. On top of that, sometimes whole manufacturing runs aren't up to snuff - this has even happened to IBM, and many Asian manufacturers have been hit by bogus parts. Burn-in and QA only catch gross errors.

jg

--
@home.com is bogus.
http://www.signonsandiego.com/uniontrib/20060323/news_1b23chinatax.html