Among the infinite-seeming decisions IT teams must make, data center hardware selection is one that's more philosophical...
than technical. Is it better to build your own hardware or to buy a plug-and-play cluster? It's no simple matter. Given all the hype surrounding converged and hyper-converged systems, this fundamental question takes on interesting new complexities.
Buying a la carte prevents vendor lock-in and purchases are based on whether the product works for the specific job. This approach, of course, requires some level of system integration. It also adds time to the implementation process.
On the other hand, buying a prestructured cluster can reduce the time needed to get systems running while easing some of the management burdens. The cost of acquisition, however, can be a hurdle.
Find the right approach to choosing data center hardware
The epitome of the prestructured strategy is the containerized data center, in which a vendor packages a huge amount of pretested gear so it can be delivered and installed quickly. If your IT team lacks hardware integration skills, the plug-and-play approach -- whether it involves an entire data center in a container or a converged infrastructure product -- will be attractive to an organization that needs reliability and scalability.
First, the pieces are guaranteed to work together. All of the components are accounted for, and a basic setup guide is typically all users need to get started.
So, do those claims hold up to scrutiny? With the high level of standardization in the industry, described generally by the term commercial/commodity off-the-shelf, or COTS, we are close to truly interchangeable elements. Modern systems are akin to Legos with interlocking parts. This dispels some of the supposed magic of a converged system: the need to buy a converged product to get all of the parts to easily work together.
Enterprise IT should also look at claims of tuning and optimizing the converged architecture with skepticism, since that "tuning" may be for a standard use case that differs from your organization's needs. Moreover, configurations are not as flexible as some would like. Most converged clusters are relatively inflexible in configuration, especially in drive-to-system ratios and LAN ports per server.
It's worth noting that cloud service providers, with their huge buying power, can specify the hardware selection in their containers down to the screws. More modest-sized IT organizations will be buying fewer systems and, as a result, won't have that type of clout.
There is a substantial premium for the converged system, just like its predecessor the blade system. If that premium is considered an "integration cost," it allows a fair assessment against doing the integration in house or having a local third-party integrator assemble the components.
A reliance on the vendor's products brings up a thorny issue. The industry is shifting to inexpensive servers from the same vendors that are producing gear for the cloud providers. With sales in the millions of units each year, these products match the quality of anything on the market, but are available at a much lower price.
Another challenge to the converged approach is coming from the software-defined movement. With its emphasis on simple, bare-metal switches and storage, software-defined infrastructure should resolve any remaining integration issues, since elements included in the software-defined ecosystem must adhere to intercommunication standards. This would seem to weaken the argument that the pre-integration of a converged product is worth the extra expense.
Make your case
So, when does a single-vendor approach to data center hardware make sense?
If you don't want to bother with integration, a converged infrastructure product is appealing. This is especially true in building clusters for ruggedized needs, such as for the military or the oil and gas industry. But, in most cases, the real issue is the value and price of pre-integration.
Using original design manufacturer (ODM) boxes as a baseline, businesses can benefit by pre-integrating and testing the units. Likely, value-added resellers and integrators will provide that value to purchasers. The extra cost is offset by convenience, expertise and quality of process. Putting a bunch of drives into carriers and building up a box is tedious work, and not every IT team is going to have all the debug tools needed to test the units.
This is the justification Google uses to buy big preconfigured products. This same argument could also apply to a business operating only a small cluster of units.
Many ODMs offer a pre-integration service for their converged systems. ODMs tend to be flexible on items such as drives and add-in cards -- avoiding after-market lock-in the way you might be when committing to a single vendor.
It's worth noting pre-integration is nothing new. Blade servers aimed to address the same issues, but inflexibility and high costs kept them from being widely adopted. And while converged products are a better play than blades, similar issues are present.
In the end, pre-integrated systems will be attractive for companies with little interest or skill in hardware integration. They'll also be good for those with a preference for working with one vendor rather than many, or for companies that want a quick turnaround from the time they purchase to the time the systems are operating.
How this will play out in a market increasingly dominated by ODMs is still unclear, but value propositions will change dramatically -- and CEOs still tend to consider cost as king. This will work toward ODM component-level purchases for a while, but ODM converged systems could change that equation.
Server buyers may not be who you think they are
What you need to know about white box servers
Converged infrastructure born from buy vs. build