[SATLUG] Increasing speed

[Al Castanoli]

On Sat, 2007-07-21 at 23:18 -0500, tom weeks wrote:
> On Friday 20 July 2007 01:09, Brad Knowles wrote:
> > On 7/19/07, Geoff wrote:
> > >  probably a simple case of semantics.  You KNEW what he meant, even
> > > though it wasn't the 'technical' term.

> > When you use a misleading term like "RAID-10", which doesn't
> > officially exist, you need to be careful.

> > I know of several instances where people have not understood the
> > distinction between RAID-0+1 and RAID-1+0, and why you might want to
> > use one particular type over the other, and they lumped all such
> > concepts under the term "RAID-10".

> No one that I work with "lumps the concepts"  RAID-0+1 and RAID-1+0 under the 
> common term RAID-10.  Just the opposite.  We always warn customers about the 
> differences between RAID-0+1 and RAID-1+0 (although we do call that latter  
> RAID-10).

I agree with Tom and Brad in this discussion. General usage of "data" as a 
singular noun rather than plural, similar to "deer" doesn't change the fact 
that data us the plural of datum, just like media is plural for medium.  
General usage doesn't make it right to use media as a singular.  

While there are defining standards for RAID 0+1 and RAID 1+0, which make 
their implementations "standard" between all manufacturers that sell RAID 
setups with these names. HP's, Sun's and EMC's implementations of RAID 6
or RAID 10 may be completely different from their competitors' setups.
 
When planning Veritas Clusters on a multi-Terabyte disk array attached
to 72 UNIX and Linux servers a few years ago, we had HP, EMC, Sun, and
Oracle pre-sales engineerings agree that part of the project we were
planning should be on RAID 1+0. They all knew what standards that
entailed, but nobody tried to call that RAID 10. You can call that
symantics if you wish, but it's important to have all parties on a large
project like that (which now supports over 3000 active commercial web
sites) understand the meanings of technical terms.

> > If you're going to use a non-standard term, then you need to define
> > what that non-standard term means in terms of standard terms.

> I agree with your rant in theory.. but not in practice.
> I would say that RAID-10 is an implementation level term that has fallen into 
> common use... into modern de facto use.  

In current production I have a few Terabytes of storage on an HP disk
array connected via 4Gbps fibre channel to an 8 CPU Sun server, then
NFS4 mounted to other 2, 4 and 8 CPU Sun and Linux servers. The disks on
the HP array are all set up as RAID 1+0 and I daily back up 50-60 GB
volumes from the disk array to an external SCSI tape. I may have been
able to set all this up using RAID 10, but I'm not sure all the various
vendors involved would have played well with HP's implentation of RAID
10, whether that is a modern de facto term or not.

> I think that the big difference here is to educate and point out the 
> implementation, cost and statistical fault tolerance differences between 
> RAID-0+1 and RAID-1+0.
> 
> This is a good starting place:
> http://en.wikipedia.org/wiki/Nested_RAID_levels

Al "admittedly a retired linguist" Castanoli