buchachon - Fotolia

Kaleao sticks an ARM into hyper-convergence

Kaleao gambles that customers will buy its KMAX hyper-convergence; it favors 64-bit ARM processors and microvisors, rather than x86 hardware and traditional hypervisors.

Newcomer Kaleao is moving into the hot hyper-converged market with a twist, eschewing x86 servers and mainstream hypervisors.

The startup today launched its KMAX hyper-converged platform, combining compute, flash storage, virtualization and networking in one box. KMAX is built on 64-bit ARM processors instead of the Intel x86 technologies commonly found in hyper-converged systems.

ARM processors are typically used in consumer mobile devices, although Hewlett Packard Enterprise (HPE) uses ARM for the controller in its StoreVirtual 3200 low-end storage system.

KMAX appliances and servers will be available via an early adopter program this month. General availability is set for January 2017.

Based in Cambridge, U.K., Kaleao claims its ARM-based systems use fewer than 15 watts of energy for each eight-core server. Along with being more energy-efficient, Kaleao marketing manager Giovanbattista Mattiussi said ARM also allows greater innovation.

"Intel has great technology, but comes with a lot of constraints," he said. "The server board for Intel is similar to a PC board, but bigger. Innovation is limited. With ARM, we had the possibility to redesign this architecture. We came out with something that imitates convergence, with software, network and compute residing on the same physical board, but, at the same time, they are accessible separately. So, we are also disaggregated. We call this disaggregated convergence."

KMAX building blocks consist of appliances and servers. The hardware is the same, but appliances include embedded OpenStack software, software-defined networking and storage features, such as replication, snapshots and multi-tenancy.

Kaleao uses microservers instead of traditional hypervisors in its hyper-converged systems. Kaleao calls its hypervisors microvisors. They do not require a guest operating system, and they orchestrate hardware and software resource sharing across pools of KMAX storage. Mattiussi called the architecture "physicalization, a physical virtualization."

"Via the software, you can set up a virtual machine associated with a four-socket server with a certain amount of memory and a certain amount of storage," he said. "The memory can come from anywhere in the cluster -- same thing for the storage."

Kaleao also integrates networking to a greater degree than most hyper-converged systems.

KMAX blades -- each consisting of four compute nodes -- connect through embedded 10/40 Gigabit Ethernet switches. Each blade has two quad small form-factor pluggable Ethernet ports on the front panels. The embedded switch sends networking packets across compute nodes.

Guest machines are assigned to network adapters when created. When application domains are created, network adapters are dynamically assigned a unique MAC address, with routing policies pinned to the domains.

Each server or appliance includes an eight-core processor, 4 GB of main memory and up to 128 GB DRAM cache. For storage, each compute node supports four NVMe SSDs. Each SSD can be 500 MB, 900 MB or 1.9 TB. Four servers go into a node, and four nodes -- 16 servers -- make up a blade.

A 3U chassis holds 12 blades for 192 servers, 1,536 CPU cores, up to 370 TB of SSD capacity and 960 Gbps network bandwidth. A chassis can also be stacked in 42U racks. Pricing for a server blade starts at under $12,000.

Kaleao takes a risk by going against the hyper-converged grain. Hyper-converged products have caught on largely because they include commonly used Intel hardware and mainstream hypervisors, usually VMware. Newcomer Kaleao is trying to rewrite the rules to take on the likes of Nutanix and VMware VSAN-based products from Dell EMC, HPE, Lenovo and others.

Henry Baltazar, storage research director for 451 Research, calls Kaleao "the Tesla of hyper-convergence" partly because of "the weird, interesting things that they can do density-wise and power-wise."

Baltazar said Kaleao makes sense for use cases such as cloud and internet of things applications.

"They'll get a look for some next-gen use cases," Baltazar said. "There's tons of flash in there, and it's priced aggressively. They also have substantial network interconnect between nodes and containers."

He said Kaleao will have to prove itself by lining up big customers early for its technology to catch on.

"You can build it, but will [customers] come?" he said.

Next Steps

Hyper-converged market explosion explained

Converged vs. hyper-converged: Where to begin

Why HIC needs more user testing, validation

Dig Deeper on Hyper-Converged Vendors and Products