Edge computing is a hot application for HCI. These environments typically require smaller footprints and lower power consumption. HCI vendors have been introducing “Edge Nodes”, smaller capacity, lower core-count CPUs and lower-cost models to address this new market. But most are fairly similar to their other models – maybe 1U instead of 2U and a single CPU with fewer cores, etc. Scale Computing’s HE150 is much different.
The HE150 runs on the Intel NUC 10, measuring 4” x 4” x 1-1/2”. And while much smaller, it runs the complete HyperCore OS, same as the other HC3 HCI models, not a stripped down version of Scale’s software. This includes automated tiering, remote replication, writable snapshots, WAN optimization and thin provisioning, plus a VM migration utility.
Currently, HE150 offers single-CPU options with 2-, 4- or 6-core configurations. Each NUC has up to 64GB of RAM, an M.3 NVMe SSD with up to 2TB of capacity and a 1GbE connector. This enables a single connection between nodes, but future models will have multiple options for Ethernet and Thunderbolt connections. According to Scale Computing, a cluster of three HE150 appliances can cost as little as $5000.
Scale Computing has been in the hyperconverged infrastructure space for number of years, carving out a successful business selling HCI systems to SMB and mid-market companies. The software-defined storage layer for this HCI product was originally developed as a scale-out storage system. Scale was one of the first companies to define hyperconverged infrastructure, adding compute to their scale-out architecture in 2012 – although the name HCI came later.
Scale has designed this product with IT generalists in mind – or non-IT folks in mind. Management is done at the VM level, meaning users don’t need to be storage admins. They also include a KVM-based hypervisor with the system, in fact, it won’t support VMware or any other hypervisors. This could be a limitation for some companies considering this product, but for their target market, many of whom were taking their first foray into server virtualization, it was a benefit. Eliminating the hypervisor license was also a big cost savings.
A few years ago, Scale was chosen as the HCI solution for the EMEA grocery chain Delhaize that wanted a standard IT platform for its retail stores, beating out some other well-known HCI vendors. Lenovo partnered with Scale Computing on the hardware, which was eventually released as Scale’s HE500 series, single-CPU nodes in 1U and tower configurations designed for small spaces and lower cost.
Scale Computing’s Hyperconverged Compute Cluster (HC3) system consists of three to sixteen 1U nodes, each running HyperCore, the Scale Computing software at the Linux kernel level (not as a VM), that handles cluster management. SCRIBE (Scale Computing Reliable Independent Block Engine) is their backend data placement engine that parses data into blocks, makes a minimum of one additional copy, then distributes blocks across all drives on all nodes to maintain redundancy and eliminate single points of failure. A Scale cluster can support up to 16 replicas, designated by the VM.
They present these data blocks directly to the hypervisor, eliminating the multiple layers of file and block translation that occurs in other systems as data moves from physical disks through the virtualization layer (which is essentially another file system), then VMFS and the file system on the guest VM. The hypervisor is controlled by the system, the system doesn’t run in the hypervisor. Management of cluster is done from any web browser, logging into the IP address of any node.
Scale Computing has developed a software-defined networking (SDN) overlay connecting a cluster of HE150 nodes into a ring topology enabling node-to-node communication and data transfers without a switch. Using BGP routing, a custom daemon and APIs, the Edge Fabric provides redundant connectivity between nodes using a single NIC and allows each node to connect to an outside network, which can also be used by other nodes. There are performance limitations, compared with traditional switched fabrics, but for edge environments that need a resilient, low cost compute platform, the benefits of a direct-connect topology are obvious.
Scale has always touted their software stack as being thinner than other HCIs, requiring only a few GB of memory to run. This lowers the resource requirements of a Scale cluster, keeping costs down and supporting their target market of smaller companies. But the efficiency of the HyperCore architecture also makes this an ideal software to run on micro-servers, like Intel NUC. According to Scale Computing, they have optimized their HC3 OS, enabling it to run on 1GB of memory, leaving more resources for workloads.
This is the first HCI vendor to come out with an Intel NUC model and, based on Scale Computing’s existing technology, it seems like a good fit. Scale has historically done a good job of focusing on the small and mid-market company space, designing their products to support this choice. Including the hypervisor and leveraging software-defined networking technology in their Edge Fabric, they have kept the cost down while providing the resiliency required to use HCIs in large enterprises – albeit in small deployments. With a 4” x 4” footprint this is an HCI that can fit under the counter and still provide full VM-based functionality.