Hyperconvergence was supposed to be nirvana for the data center. Converge compute, storage and networking into a single box and all data center problems go away. The problem is that didn’t happen. Hyperconvergence is the ultimate white board technology: It looks good on the board, and even looks good at the start, but as it scales, problems start to arise.
There are three distinct hyperconverged models; Hardware Hyperconvergence, Software Hyperconvergence and hybrid hyperconvergence.
Hardware hyperconvergence is a solution where the vendor provides everything; server, internal server storage, and network components. It is also the ultimate form of vendor lock-in because all the nodes have to come from the original vendor. Most of these solutions also do a less than adequate job of integrating existing storage resources. For all intents and purposes, it is a data center reboot. If the organization is small or if there is a new project underway then this type of solution is probably acceptable as long as the vendor has long term viability.
Software hyperconvergence, while potentially more difficult to implement, is more flexible. It allows the organization to select the hardware and internal storage it wants or leverage existing hardware. It also, typically, allows the mixture of node types.
In both cases the network becomes the key concern. East-west traffic (communications between servers) is already a challenge in standard virtualized infrastructures. A hyperconvergence infrastructure that combines the compute and storage traffic only compounds it. As the hyperconverged infrastructure scales, IT planners need to be very careful with what type of networking components they deploy and they need to carefully monitor its performance.
The third type of convergence is more of a hybrid. Most converged infrastructures are a cluster of servers where storage is aggregated across nodes into a single pool. That means a simple write will involve most if not all of the nodes. In a three node hyperconverged cluster, the network traffic is not significant but when nodes increase to a dozen or more, the network becomes critical.
A way to defuse this problem is to create a hybrid hyperconverged architecture. In this architecture, read IO stays local to the node where the virtual machine is running, write IO is sent both to the local node and to a centralized storage system. The centralized storage system enables both data protection and virtual machine mobility, but eliminates much of the east-west IO.
Which is Best?
Which is best depends largely on the state of the data center. There is no denying the ease of implementing a turnkey system. But most software-only solutions have relationships with hardware providers in which their software is preinstalled and software solutions provide the ultimate in flexibility. Hybrid solutions deliver much of the flexibility and reduce east-west traffic but do incorporate a little old school to do that.
By and large, Storage Switzerland recommends ruling out hardware systems so organizations can avoid vendor lock-in. The selection criteria then comes down to how confident the organization is in its east-west network vs. the downsides of a centralized storage system.
To learn more about hardware-based and software-based hyperconvergence watch our webinar, “Showdown: Hardware-Based Hyperconvergence vs. Hyperconvergence Software”.