Every so often it makes sense to take a step back and attempt to define the terms that we in the industry use to describe the approaches we take to solving customer problems. One of those is “converged architectures”. When defining anything in the IT industry you have to strike a balance to make sure that the definition is specific enough to be useful, but not so detailed that it becomes a product definition instead of a category definition.
The Problem that Convergence Claims to Solve
Converged and hyper-converged infrastructures claim to make life simpler for the overburdened IT staff, with spending less time implementing and managing the virtual infrastructure. Some of these solutions, especially hyper-converged, also claim to reduce acquisition costs. A simpler, less time consuming and less expensive environment certainly sounds appealing.
The Types of Convergence
We break the converged infrastructure category into three subcategories: hardware converged, turnkey converged and hyper-converged. Hardware convergence is the pre-integration of existing, independent hardware solutions. We will detail this subcategory in this article.
Hyper-convergence is a software-only solution that runs alongside or within the hypervisor cluster. In this solution you have maximum flexibility picking the storage and servers that you want but you may increase complexity.
Turnkey converged pre-installs the above hyper-converged solutions on either off-the-shelf or customer-supplied hardware. They hope to offer a balance between the simplicity of hardware converged and the cost savings of hyper-converged. We will detail both hyper-converged and turnkey converged in a future article.
For us, hardware convergence is the pre-integration of independent hardware solutions, meaning that each component of the converged stack is already a stand-alone product. Often these are name-brand vendors partnering to deliver a more turnkey appliance. Interestingly, two of the components are common to most of the converged solutions. The hypervisor is almost always VMware. That is not a surprise given VMware’s market share. What is a surprise is that the server component is almost always Cisco UCS. Clearly there are some political reasons for this widespread inclusion, but there are also some technological reasons. We will explore the Cisco angle in a future blog.
At first glance this type of convergence seems more like marketing than engineering, but in conversations with data centers that have adopted this strategy this type of convergence has made a difference. In most cases it reduces implementation time, is optimized to deliver better performance and makes support issues easier to resolve.
The hardware converged solutions are more than just a reference guide, in most cases they are a single SKU that the customer orders. The participating vendors go to great lengths to make sure that the components are optimized to work together, and in most cases they provide detailed runbooks of the solution.
What is lacking in most cases is real software integration. Clearly some of this is delivered via the independent vendor’s vCenter Plug-In integration. But we would like to see more than that, and I think we will eventually. For example, there would be value in setting quality of service (QoS) levels one time that would adjust CPU, network and storage resources in one fell swoop.
The other downside to the hardware converged approach is that it’s doubtful these solutions are going to be less expensive than the other converged subcategories. These are typically name brand hardware solutions running on a name brand hypervisor.
Hardware converged architectures reduce the time required to stand up the first VM and should reduce support issues as well. Also, as we start to see hardware converged architectures in which the storage component is an All-Flash array, the pre-tuning of the network component by a flash expert should lead to better flash performance and even more dense virtual infrastructures. That combination should provide a rapid ROI for these environments.
In my next column we will cover the opposite end of converged category, hyper-converged architectures. They should dramatically reduce the cost of virtualization, but do they increase its complexity as well?