On Tue, Mar 06, 2007 at 03:11:51AM +0000, Martin wrote:
> On Mon, 2007-03-05 at 20:24 +0100, Herbert Poetzl wrote:
> > On Mon, Mar 05, 2007 at 07:43:51AM -0500, Chuck wrote:
> > > On Monday 05 March 2007 06:15, Herbert Poetzl wrote:
> <snip>
> > >
> > > controllers. i would rather see the boss change the case to a 2u and
> > > put a real hardware raid controller in on a 2 card riser but...... it
> > > is not my call.. (and of course we find all this out after the machine
> > > has been in our production environment for 5 months)
> >
> > in most cases the hardware raid controller is not worth
> > the money, as a software raid usually gives a much better
> > performance with less latency and more control for the
> > operating system ...
> >
> > nevertheless, hw-raid can have some advantages if it is
> > done properly, e.g. auto reconstruction without affecting
> > the system performance and/or battery buffering in power
> > failure cases ...
>
> I used to like the idea of hardware RAID but two things put me off:
>
> 1. When you pull the power on a system apparently the memory goes
> first but I/O systems function for just a bit longer - often writing
> junk data. This is apparently one of the things the high end UNIX
> vendors used to spend money on trying to get right. In short, you
> *need* a battery backed hardware RAID if you are serious about
> avoiding data corruption. These are more expensive. It also makes any
> form of RAID device that requires drivers to run (i.e. the soft-RAID
> devices on many modern machines) a little questionable to my mind.
>
> 2. Data corruption is serious because none of the formats the hardware
> RAID systems use are public. I am under the impression that in many
> cases even data recovery specialists do not have access to these. Thus
> you are completely at the mercy of the tools the vendor gives you. If
> they are buggy or you get into a situation (see above) that they can't
> recover from it's game over.
>
> Thus, I would *strongly* advise that unless you /need/ the performance
> a hardware RAID controller gives (and can then afford the UPS and the
note that the 'performance' in many cases is a myth,
for several reasons, mainly because:
- hardware raid has 2-256MB cache, software has 1-4GB
- hardware raid has a single channel to the host,
while proper setup soft raid can burst over N channels
simultaniously (and will do so, e.g. for separate I/O
threads)
- elevator in the kernel, vs limited TCQ
best,
Herbert
> high level service contract with the vendor, etc.), use the Linux
> software RAID.
> If it all goes wrong you can always read the source and piece things
> together manually. I've had to do this. It's not fun but it is
> possible. For me it made the difference between having to tell my boss
> that the fileserver would be down for a while and having to tell my
> boss that we would have to revert to last months backup.
>
> HTH
>
> Cheers,
> - Martin
>
>
> _______________________________________________
> Vserver mailing list
> Vserver@list.linux-vserver.org
> http://list.linux-vserver.org/mailman/listinfo/vserver
_______________________________________________
Vserver mailing list
Vserver@list.linux-vserver.org
http://list.linux-vserver.org/mailman/listinfo/vserver
Received on Tue Mar 6 15:20:52 2007