Yael McGuire works with Facebook on the Open Compute Project. He made a point in a session today at the Intel Forecast event that is worth exploring.
His remarks illustrate the difference between companies like Facebook and EMC and how they view the ways data center infrastructure should be treated.
McGuire countered the view about converged infrastructures in data centers. Instead, he said, Facebook thinks of the future more in terms of a diverged infrastructure.
A converged infrastructure means that the storage, network and compute is integrated into one box. A diverged infrastructure means a horizontally, scaled out infratructure.
“It’s the Web way or the highway,” said Adron Hall with Tier3, here for the Intel event.
Hall reflects what you see here today at Intel Forecast, the inaugural event for the Open Data Center Alliance. The event highlighted the difference between constituencies that have different DNA. Hall represents the clouderati community who see the world much differently than the legacy software providers, who possess a DNA rooted in on-premise deployments. Some are starting to adapt but most are clouding terminology to bolster their core business. IBM, SAP, HP and Oracle all do this. Most of those attending Intel Forecast reflect a community that have depended on these vendors. They have been in IT for most of their careers. For the most part, they have established a core competency in virtualization but the next step forward is one that you can tell they are unsure about taking.
Virtualization is different from a scaled out cloud infrastructure. You can virtualize your infrastructure but it’s not the same as taking advantage of a platform-as-a-service (PaaS) to develop new apps and deploy to a scaled out, infrastructure as a service (IaaS) such as Amazon Web Services.
Following McGuire’s analogy, a scaled out cloud infrastructure becomes a living, breathing organism. The organism still lives even if one node or a group of nodes dies. It can handle data from multiple sources. That’s true of AWS. It simply offers a data fabric. It’s a service for app providers, SaaS companies, PaaS environments and a host of other offerings.
A converged infrastructure is limited by the physical constraints of the box itself. It does not scale out horizontally. It can handle data from multiple sources, depending on the applications that are part of its infrastructure. The solutions offered through VCE, for example, are quite popular. Their use cases are different than what you will see from AWS. Instead, VCE will run solutions such as SAP on Oracle, which it is primarily suited for.
The converged infrastructure v. diverged infrastructuree debate is not one about which is better. It’s simply a matter of what suits you best.
I spoke to one Intel executive who said it well when she described it as workload affinity. She means that a mission critical workload may require different requirements than a SaaS. By measuring workload affinity you can define your infrastructure requirements.
I think that is true to some extent. But for pure value, I’d be looking at what business I want to have and the role that services play in that strategy. Is it diverged or converged?
My bet is that most businesses will look at a diverged infrastructure for the ease of which it is to deploy across a service and the capabilities it provides for management.
Why? It’s just really hard to beat commoditized technology.