There's an article on DevX today entitled 'Why You Need to Consider a Single-System Image.'  With all due respect to the author, I think his advice is dated.  Ten years ago, this was true - if you wanted a grid you had best focus on a single system, all components in lockstep.  This is not true anymore, and the tools available to us today allow us to address the concerns he lays out in his article. 

 

The end-user often presents the following resistances to grid computing projects: 

  1. Monetary – people like the idea that what they paid for is not being used for someone else
  2. Political – sometimes folks just don’t want to be part of a larger and organized infrastructure that will force them to share the success and failures of others</meta>
  3. Technology – proper technology is not in place to allow such infrastructure to be built</meta>

 

I would contend that with proper technology, the first two are completely meaningless.  Today, utilizing tools like OpenStack, or VMware vSphere, you can provision a collection of servers to many people, and provide them with the levels of service they want, all while hiding the costs and utilization of the hardware, and letting them manage the platforms of their choice. 

 

In an old-fashioned grid world I would have to choose to deploy my Oracle applications on a particular version of the system image - and then I have to pray that subsequent versions of Oracle do not outpace the evolution of my system image.  It would be horrible, for example, if my system image were based on RH 3 and I had to use Oracle 11gR2?  Or it was based on RH5 and I found a need to use Microsoft's SQL server?  Simple, I'd be hosed. 

 

No, the true answer to grid computing is what we're seeing evolving in the marketplace - the complete domination of the hypervisor.  You build your applications on whatever platform you need, you deploy them into this cloud of servers (ESX, XEN, Amazon EC2, Rackspace) and provision them to your end-users, or let them do it on demand themselves.  There is a growing storm of tools coming to the marketplace to allow self-service in the cloud.  Amazon EC2 is by definition self-service.  OpenStack.  Surgient and Lab Manager today in the VMware area - vCloud Director in the coming months.   

 

The future is the hypervisor - a day is coming when we don't even install applications anymore - they are managed on-premises by vendors, or managed in vendor clouds that talk to our own.  SalesForce.com was pretty much the first successful SOA vendor to adopt this model.  Microsoft, Google and others are scrambling to get a place at the table in this new world. 

 

While I'm not bullish on off-premises cloud offerings, I nevertheless understand the benefit they can have for offering low-cost resource utilization when demand is high.  But on-premises grids are HERE, today, and the hypervisor has made it all possible.  No need for special system images when quick templates off of VMTN or straight from SuSE or RedHat serve just as well.

 

Our expense in managing the server closet usually isn't the servers themselves, but the very expensive software that they run.