VMware's EVO:RAIL product launch at VMworld 2014 has certainly helped spur along the idea of implementing a hyperconverged...
infrastructure in the data center. A hyperconverged system takes the disparate pieces of compute, networking and storage, and ties them together.
What is less clear to many potential buyers is the value of hyperconverged infrastructure. A hyperconverged infrastructure builds a cluster of commodity servers into a vSphere cluster without needing a storage array. Simplicity is a guiding principle: Simple to deploy, simple to manage and simple to scale. This new architecture is a very different way of building an infrastructure for virtualization. Before adopting hyperconverged infrastructure, it is important to understand what it is and how the parts work together.
Commodity servers at the core
One of the first things to notice about hyperconverged is that the physical servers are standard x86 servers. One hardware vendor's servers could be swapped for another without changing the architecture. For example, EVO:RAIL will ship from several hardware vendors. Other examples are the various partnerships, SimpliVity with Cisco and Nutanix with Dell, that use branded vendor servers as a platform for hyperconverged infrastructure. Naturally, a good quality physical server is the foundation of a good virtualization platform, but the actual x86 server is not the magic here, it is a commodity.
Merging storage in the servers
One of the defining features of hyperconverged is that inside each of the commodity servers is some storage. The hyperconverged platform takes the local storage in a group of servers and makes it into redundant, shared storage. Rather than a storage array as a separate box, hyperconverged infrastructure uses the same commodity servers that provide compute.
Some of the magic is in the software that takes the islands of local storage in each server and joins them as a storage pool. These offerings cluster the nodes to provide redundancy, removing the risk that VM storage would be lost if a physical server fails. The cluster doesn't use disk-based RAID and LUNs to manage storage. Rather redundancy and performance are service levels that are configured for each VM stored on the cluster. VMs stored with different policies will be laid out across the cluster in different ways to provide higher or lower redundancy and performance.
Be aware of the pros -- and cons
To add capacity to a hyperconverged infrastructure, you add more commodity servers. As you supplement servers for more compute, you simultaneously put in more storage capacity and performance. This coupling of compute and storage growth is the key to scaling out to build a large infrastructure. Some products let you add as little as one server at a time, allowing you to size your infrastructure for today's workload. This avoids one of the classic problems of sizing storage for three to five year's of growth -- but paying for it in year one. Spending on hardware can now be a quarterly or annual incremental cost, rather than a huge expense every three years.
Keep in mind is that each x86 server may not scale up as far as a standalone server. There are server models that support multiple terabytes of RAM; current hyperconverged platforms usually top out at 512GB of RAM. Similarly, hyperconverged usually has only two CPU sockets per host and may have limits to the disk performance that can be delivered to one VM. Be aware of these limits and their potential impact on your environment before adopting a particular hyperconverged solution.
Networking in converged systems
Most hyperconverged offerings involve a single network. The storage traffic is carried over the same 10 Gigabit Ethernet (GbE) network as platform management and VM traffic. This simplifies data center cabling and means there are fewer technologies to support. Most products do allow the storage to be run over dedicated 10 GbE networks with VM traffic on separate physical ports. The 10 GbE network is what joins all the nodes in the cluster and allows storage replication for redundancy. Readily available 10 GbE has been a key enabler for hyperconverged -- all the vendor offerings won't work without it.
A simpler way to administrate
One of the cornerstones of hyperconverged infrastructure is that management should be quick and easy. Often all of the management is unified into the vSphere desktop or Web client, a familiar place for vSphere administrators. The storage architectures of hyperconverged solutions also make storage management simple with no LUNs or RAID groups to setup and no storage fabric to configure.
If VMs need specific storage configuration, then it is usually a policy that is applied to the VM. The underlying clustered storage provides the capabilities at VM granularity. The simplified management allows companies that deploy hyperconverged infrastructure to move on from thinking about this infrastructure. Most IT departments need to do more with less. The less time they spend deploying and managing their infrastructure the more they can focus on applications and business needs.
Hyperconverged is different but the same
Hyperconverged infrastructure is fundamentally different to the silo-based infrastructure that it replaces. The simplicity of deployment and growth is very attractive to IT teams struggling to balance the future demand against the current budget. The ease of setup appeals to IT teams that wish to focus more on the applications and business needs. As with all products you deploy and rely upon, understand the benefits and drawbacks of the solution before you commit.
How EMC and Dell plan to sell VMware EVO:RAIL hyper-converged appliances