The IT management industry remains in a white-hot frenzy over OS virtualization. The introduction of the hypervisor has enabled new capital and operational cost models that just weren't available before. And the technology is changing the efficiency dynamics of data center management by orders of magnitude.
The abstraction of the software world has altered things: By inserting a hypervisor, software stacks can be consolidated, migrated, failed-over, life-cycle managed, and defined logically in all kinds of other new ways. That leads to both better capital efficiency and better operations efficiency.
Everybody grasps these points because almost everybody understands software. So then, why are data center administrative costs still rising?
My theory: It's because of some of the non-software parts of a data center that still need simplification, management, and automation.
I'm talking about the "forgotten half" of the data center: Physical server I/O. Physical networking/cabling. Physical switches. Physical load balancers. Storage connections. Even with OS virtualization in place, configuration changes still force your admins to walk the data center floor with cables, screwdrivers, and I/O cards to reconfigure or repair these items.
And the most sobering observation is that OS virtualization actually doesn't solve these physical issues. My contention is that the infrastructure half of the data center is what's contributing to the ever-growing portion of data center costs: management and administration (see chart).
Source: IDC, Transforming the Datacenter: Consolidation,
Pervasive Virtualization, and Energy Optimization
Now, consider a new concept called converged infrastructure. You may have heard the term, or you may have heard of unified computing (the other way of referring to it). If not, you should familiarize yourself with the concept, because it's coming to a data center near you. And it's going to change the efficiency and economics of how the "other half" lives.
Converged infrastructure technologies are analogous to OS virtualization -– except they're applied to the non-software half of the data center. They include:
- I/O virtualization consolidates logical I/O devices onto a single I/O device, which can be logically configured into any type or quantity of things like NICs, HBAs, CNAs, etc. This obviates the need to change and configure physical I/O devices. It also allows admins to define and/or clone a physical server state (addressing, ports, worldwide names) at will.
- Network convergence consolidates network traffic onto a single wire. This wire can logically carry traffic from multiple networks and data streams, and it can be reconfigured nearly instantly. This eliminates the need for multiple (and frequently underutilized) cable runs, resulting in a "wire-once" environment.
- Virtual switching and IP load balancing. When used in concert with these other technologies, this allows IT operations to define any logical network topology on an at-will basis.
Now, combine these three technologies with an integrated management console, and presto -- you have essentially virtualized the infrastructure of a data center.
OK. Nice technology -- but what about the economics? Well, like OS virtualization, converged infrastructure makes a huge dent in the remaining inefficiencies in the data center:
- Capital. Like OS virtualization, converged infrastructure helps consolidate use of capital. In this case, that means more efficient use of network and I/O assets with fewer cables and network components. It frequently also results in fewer (or at least less expensive) I/O components, which can range from $100 to $1,000 apiece. This is because a single physical I/O component can be made to logically represent many physical I/O components, which no longer need to be purchased.
- Operations. Converged infrastructure can eliminate many of the operational steps needed to provision servers, software, and infrastructure. It also allows IT staff to work more efficiently by provisioning infrastructure items logically rather than physically. All in all, I've seen this approach cut the number of run-book steps required by up to 75 percent.
My contention is that to contain -- and maybe even reduce -- data center administrative costs, we have to address the efficiency and automation requirements of the non-software half of the data center. And as data center operators find that OS virtualization gets them only partway to true efficiency, they'll see converged infrastructure as the next great opportunity.