Update March 2017

With the advent of network virtualization and hyper-converged infrastructure (HCI), the nature of network traffic is undergoing a significant transformation. A dedicated Fibre Channel network with many of the advanced storage features located on a dedicated storage array is no longer the norm. Storage and application traffic both can reside on Ethernet networks with many advanced storage features executing in a distributed fashion across a network. 

In the early days of software-defined storage and HCI, when most of the workloads were CPU-intensive or memory-intensive, the network was often an afterthought. The increasing I/O and storage-intensive workloads within software-defined storage/HCI solutions, place stresses on the network that ultimately impact performance on inappropriately designed networks. Gartner, in recognition of this trend, forecasts 10% of hyper-converged solutions suffering from unavoidable, network-induced performance problems by 2018. Gartner further predicts 25% of hyper-converged solutions being plugged into the wrong tier of the data center network in that same timeframe[1].

As such, the overall network design plays a larger role in defining the performance of software-defined storage/HCI-based solutions and the applications that run on them. The goal of this document is to identify best practices for the software-defined storage/HCI solution from EMC ScaleIO (ScaleIO). This best practice guide not only mitigates network-induced performance problems, but also sets the foundation for a network with scalability engineered from day one. This provides a stable and predictable foundation for the future growth of any software-defined storage/HCI solution.