[Ma-linux] Cloud Computing

[Chris Rogers]

...and the second response.

Serge Wroclawski wrote:
> Finally a technical discussion!

I agree.  If only from the standpoint that this is in my close area of 
interest... :)
> 
> On Fri, May 29, 2009 at 03:23:07PM -0400, Przemek Klosowski wrote:
> 
>> Chris likes Xen, I like KVM--we'll see. 

Again, both have their place.
> 
> I think focusing on the hypervisor is a bit like focusing on which 
> kernel scheduler we'll be using next year. It's interesting but the 
> benefits of one vs another are relativaly minor.

Yes and no.  The underlying technologies are SO different, that 
efficiencies of scale will become very apparent, especially with a cloud 
model.  Anyone trying to do cloud with VMWare will get a RUDE surprise 
in terms of licensing costs and associated hardware costs when they need 
to make up for that 10-20% overhead on VMWare's hypervisor.

> There are some dignificant differences between KVM and Xen which are 
> important, but for a majority of those folks planning a serious 
> deployment, the differences will be easily dealt with.

Agree.  Most won't notice.  If this is your first foray into 
virtualization, pick a free one and learn.

>> It'd be interesting to see how a proprietary company
>> buys an open-source based technology and merges it successfully---so far
>> I can only think of IBM succeeding in that, 
> 
> Well the Xensource product line has always been propreitary. Xen itself 
> is Free, but the management tools are not, and those tools are pricy, 
> coming in at or near the price of a server.

Free for most enterprise class features (including Live Migration), 
$2500 for Enterprise Essentials which include HA and Storage 
Centralization, $5000 for Platinum Essentials which includes 
Provisioning server (thin deployment) and Lab Manager (good for dev 
environment control).  Compare that to a starting price of $5700 for 
VMWare ESX VI3, and then add to it for some of the more useful features. 
  Oh, double the price for a four-socket machine, too... :)

> At work, given sufficient quanitity of servers, the cost of purchasing 
> licenses for the servers rivaled that of developing in house tools, 
> which then have the added benefit of integrating with an already 
> established system to deploy physical servers efficiently.
> 
> I gave a presentation on part of our design at a NoVALUG meeting. For 
> those folks who were at the meeting, the bootloader now is complete and 
> (when I have time) I'll be finishing up the "tracking" component of our 
> design.
> 
> I then hope to be able to turn my attention to the more "advanced" 
> features, such as automated deployment (ie a virtual server is put on an 
> appropriate server automatically), as well as some other possible 
> features, such as an automated server migration system that doesn't 
> rely on centralized storage, and possibly automatic migration.

I applaud the effort.  Hopefully you create a great tool.  In my case, 
we started down that road, and quickly wondered why we were trying to 
reinvent the wheel.  There were tools out there to do what we needed 
that were available right now.  Didn't have to wait for a dev cycle, and 
didn't have to pay a developer.  I can honestly say that this is the 
first time that my company used a COTS product for the reason that it 
was cheaper than developing the function in house.

Thanks for the discussion, guys.

Chris