As I’ve written about what I’m calling the “Rack Endgame”, the specter of converged infrastructure hasn’t been far from my thoughts. As others have pointed out, disaggregation of servers, networks, and storage doesn’t require a rack-sized stack; it can exist in a rack-mountable chassis and is already on sale!
Disaggregation and Convergence
Let’s get two key terms out of the way to start with.
Disaggregation removes compute, storage, and network components from their conventional shells (servers, storage arrays, and switches) and rearranges these components to meet the needs of hypervisor clusters and application platforms.
Historically, servers have included expansion slots, I/O ports, and management, but none of this is really important in rack scale. Disaggregation creates micro-servers consisting only of CPU and memory, shifting everything else off-board. The same is true of storage arrays, which are seeing their controllers move closer to compute or reduced to software. Networking is disaggregated as well, with SDN giving rise to a new controller-oriented architecture and most other functions becoming software, too.
Converged infrastructure takes all these disaggregated components and integrates them, with software and hardware working together to create a single-box solution. So far, converged infrastructure products have moved formerly-proprietary components like array controllers into software virtual appliances running on a hypervisor. But there’s no reason a converged solution couldn’t eliminate the hypervisor and containerize these elements.
Therefore, converged infrastructure is simply an implementation of disaggregation, and fits well with the rest of my Rack Endgame concept. It could happily coexist with a top- and bottom-of-rack storage solution, and could indeed morph in this direction in the coming years. I’m sure SimpliVity will run great on the Cisco UCS rack!
Read more about Cisco’s Trojan Horse
Converged Infrastructure: Why and Why Not
Converged infrastructure solutions from Nutanix, SimpliVity, Scale Computing, Atlantis Computing, and now VMware are all the rage in enterprise IT. And for very good reason: Enterprises aren’t interested in a go-for-broke cloud; they need “buyability”, integration, and supportability.
“Buyability” is much more important than most technical people would like to believe. It’s simply not realistic to expect corporations to adopt “roll your own” solutions, no matter how good they are. Red Hat made Linux “buyable” and now they’re the poster child for this often-overlooked qualification. Now, just about every piece of cloud software has dozens of companies vying to be the “Red Hat” of their respective offering!
Another interesting element of the converged infrastructure picture is the old hardware versus software approach. Nutanix, SimpliVity, and Scale chose a hardware appliance route, integrating their software on a few hardware offerings and selling the lot in a box. Atlantis and VMware are providing software that can be run on a range of supported hardware. History shows that enterprises are much more willing to entertain large hardware purchases – they’re more “buyable” and supportable long-term. But perhaps VMware’s success with vSphere will change this!
By far the biggest hurdle to converged infrastructure success is the “all-in” aspect: Buyers have to adopt the philosophy, the architecture, and the hardware to see any benefit. If they aren’t going all-in, they might as well stick with conventional server and storage components. And going all-in requires a shift for IT to become an internal service provider, an insurmountable hurdle for too many companies.
The Rack Endgame is disaggregation, regardless of the form. Converged infrastructure is an expression of the trend toward disaggregation, but it is by no means the only approach. Companies could happily run plain-jane vSphere or Windows Server on a disaggregated “Endgame” rack, or they could shift to a converged infrastructure solution that looks quite similar but exists in a single box.