Data Locality

StarWind breaks storage performance limitations and sets the new speed record

We grabbed topnotch hyperconvergence platform, added PCIe SSDs, installed powerful software on top… and set the new HCI industry high score! Find out how we came up with this crazy idea.

Hyperconvergence is a breakthrough in IT. This technology enables organizations to move away from the proprietary storage arrays in favor of architectures built out of industry-standard hardware, which makes implementation of the new technologies easier than ever before. The leading industry vendors like Intel, Mellanox, StarWind Software, and Supermicro collaborated on the construction of a modern hyperconverged platform comprised of production-ready appliances to demonstrate how hyperconverged infrastructures (HCI) can perform.


Why did we conjoin the hardware and software from these vendors? Mission-critical virtualization workloads require the highest possible reliability, uptime, and breakthrough performance. All these vendors have prominent solutions allowing to build environments for intense virtualization workloads.


Intel provided CPUs and storage for our setup. Intel CPUs are designed right for cloud computing, real-time analytics, mission-critical data processing. Intel Optane NVMe SSDs are nothing but the fastest storage to date.

Mellanox Technologies, a leading supplier of end-to-end smart interconnect solutions for datacenter servers and storage systems, provides exceptionally high performance for the most HCIs, enabling today's corporations to meet the demands of the data explosion.

Each hardware component finds its place in Supermicro SuperServers – the “Linux” of hardware platforms. Supermicro is a major developer of new technologies that provides the facilities to make all the installation possible.


To squeeze maximum performance, we used Microsoft Hyper-V. Initially, we thought of using KVM because there are numerous readings on KVM benchmarking and KVM is a performance leader after all. However, in the end, we decided to run our environment on some mass thing.

That’s why we decided to use an enterprise-class hypervisor like VMware ESXi or Microsoft Hyper-V. VMware could not provide SDK so that we could import StarWind NVMe Initiator to vSphere: the company is developing its own NVMe Initiator that is not production-ready yet. Therefore, Microsoft Hyper-V was chosen.

We decided to use StarWind Virtual SAN as software-defined storage. One more time, VMware could not provide us with an SDS solution capable of unleashing Intel Optane potential. So, we could not unleash the bare-metal performance of our storage using Microsoft Storage Spaces Direct. StarWind Virtual SAN, in its turn, just ensures the highest possible underlying storage performance. Additionally, StarWind Virtual SAN has the lowest capacity and performance footprint among the existing shared storage solutions. As a result, it enables to squeeze the maximum performance from the underlying storage. On top of that, this SDS has no HCL so it can be run on any commodity server.

Some words about the measurements

Benchmarking has three stages.

First, we measured performance of a cache-less production-ready 12-node cluster that was built according to recommendations for the traditional 2-node StarWind HyperConverged Appliance deployment. Each server was powered with Intel® Xeon® Platinum 8268 Processor and Intel® Optane™ SSD DC P4800X Series for storage while Mellanox ConnectX-5 NICs were used for interconnection

On the second stage, we measured performance of an all-NVMe cluster where every server had Intel Optane NVMe drive for write-back cache.

On the third stage, storage was presented over NVMe-oF to avoid any performance bottlenecks.


The collaboration of industry-leading vendors enabled to build production-ready StarWind HyperConverged Appliances with the fastest CPUs, storage, and networking to see the true hyperconverged infrastructure performance.

Request a Callback Gartner`s Niche Player