[sn-cg] where to fit into OGF standards

[Yoder, Alan]

To really make the point of the demo, we need
some poster material that shows how the 
economics work.  The case for cloned boot
LUNs is really pretty compelling:

 * diskless compute servers
	- no state to rebuild when one breaks
	- reduced power and cooling requirements
	- reduced footprint in blade configs

 * cloned boot LUNs
	- reduced storage requirements 
	  (~5% marginal space cost per server)
	- hypervisor-like ability to provision
	  any one of many stored OS+application
	  images

I'm guessing a real analysis will show that if
a high-end array is serving ten compute server
images from the same base image, the cost per
running instance is lower than if a JBOD were
used, and the images merely copied.  And
performance would be quite a bit better too.

What do you think?

Alan

> -----Original Message-----
> From: CHRIS.KANTARJIEV at ORACLE.COM 
> [mailto:CHRIS.KANTARJIEV at ORACLE.COM] 
> Sent: Wednesday, March 21, 2007 4:01 PM
> To: Yoder, Alan
> Cc: sn-cg at ogf.org
> Subject: RE: [sn-cg] where to fit into OGF standards
> 
> 
> 
> > The stuff we're trying to put together now doesn't
> > have anything to do with anything that OGF has
> > done, to my knowledge.  
> 
> That's pretty much my point. We need to figure out how to 
> attract interest, or stop pretending that we're trying to 
> work with OGF.
> 
> > I'd like to steer away from ByteIO, personally.  We
> > should work on leveraging the ability of storage
> > systems to do direct transfers.  To that end, a 
> > rework of GridFTP in ANSI C would be very useful.
> > The present implementation in Globus requires one
> > to install most of the known universe before using it.
> 
> If we don't provide ByteIO, we won't get anyone to pay 
> attention. That doesn't in any way mean that we should *only* 
> provide ByteIO. We can do a lot better.
> 
> But think of this as the "free candy" that we want to offer first.
> 
> And I wouldn't count on anything interesting from GridFTP in 
> the forseeable future. We can't even get the document authors 
> to pay attention long enough to finish publishing it as a standard.
>