Hyper-converged systems are modular systems designed to scale out. Discover the pros and cons to these systems.
If you're in the market to refresh your storage or server platforms, you'll run across the term hyper-converged systems. Your first questions might be: What are hyper-converged systems, and how are they different from converged systems? The better questions are: What are converged systems, and how do they differ from traditional infrastructure systems? I answer all of these questions in this introduction to hyper-convergence.
How converged systems differ from traditional IT infrastructure
Converged systems are a natural progression from traditional IT infrastructure, which are commonly silos of systems and operations. In these legacy environments, there may be separate administrative groups and systems for storage, servers, and network. The storage group, for example, handles the purchasing, provisioning, and support of the storage infrastructure. The storage group usually maintains storage system architecture and the relationship with the storage hardware vendor. The same situation exists for the servers and the network groups. The concept of converged systems combines two or more of these infrastructure components as a pre-engineered solution.
VCE and HP offer best of class storage, compute, and network converged systems solutions that are tightly coupled via software. The idea is to give "one throat to choke" for support, as well as a single vendor to support the multiple components. In the case of VCE, the server is VMware vSphere and Cisco UCS hardware. The network is provided by Cisco, and the storage is from EMC, hence the VCE acronym for the independent company. On the HP side of the equation, all layers except for the hypervisor are HP solutions. (Note: VCE supports other hypervisors, including Hyper-V and KVM.)
The more distinct advantages to these solutions are the relatively simple design for a complex infrastructure environment. Normally, it could take months to design an infrastructure using a mix of best in class technologies. Solutions such as VCE allow for quick design by provider reference architectures the company calls vBlocks. These vBlocks are sized and designed to work well for specific virtual workload sizes.
How hyper-converged systems differ from converged systems
Hyper-converged systems take this concept of convergence to the next level. While converged systems are separate components engineered to work well together, hyper-converged systems are modular systems designed to scale out by adding additional modules. These systems are mainly designed around storage and compute on a single x86 server chassis interconnected by 10 GB Ethernet. At first blush, this just sounds like a server with a bunch of storage. From a physical perspective, this is accurate; as a matter of fact, VMware's VSAN software takes advantage of a very similar architecture. We'll talk about this more in a bit.
The differences between a hyper-converged system and servers with a bunch of disks are engineering and software. Hyper-converged solutions leverage improvements at the storage controller software layer to allow these systems to scale out. The more appliances you add, the greater the performance and capacity. Instead of scaling up by adding more drives, memory, or CPUs, you scale out by adding more appliance modules.
In addition to the simplified architecture, there's a simplified administration model. The hyper-converged systems are managed via "a single pane of glass." Instead of having a set of applications and a team to manage your storage array, a team to manage virtualization, and a team to manage the server hardware, one team (or in some environments one person) can manage the complete hyper-converged stack.
There are two approaches to adopting a hyper-converged infrastructure. You can buy a pre-manufactured system from companies such as Nutanix, SimpliVity, or Scale Computing (which all make highly engineered systems for different segments of the market), or build your own hyper-converged system. Solutions such as VMware's VSAN and HP's StoreVirtual are positioned at customers looking for more control points in the design of their hyper-converged system. With a build-your-own approach, you can select your desired server hardware manufacturer and your preferred drive configuration, just as long as the hardware is supported by the software vendor.
Drawbacks to hyper-converged systems
The drawbacks to hyper-converged systems include the inability to make granular upgrades or tweaks to the system. Storage growth and performance tuning are pain points for most organizations. If a cluster runs low on storage but not compute, you must still upgrade overall compute capability by adding another appliance. Likewise, if you want to tune storage disk configuration for a particular application, this can be a challenge.
The exception is a build-your-own platform. You have the option of adding only storage or compute for the VSAN approach. A second problem is organizational challenges brought on by combining two systems that had traditional operating within siloed organizations. If a server group wants to bring in a hyper-converged system, budget challenges and turf politics comes into play. It's very difficult to get that much storage in under the radar of the storage group. From the other angle, very few storage teams want the headache of managing compute resources that are consumed by groups outside of the storage team.
The bottom line
Hyper-converged systems may bring the advantages of simple designs, decreased administrative overhead, and simplified vendor management to highly virtualized environments. Be sure to consider the systems' shortcomings before you decide to adopt hyper-convergence.