theCUBE at HP Discover in Barcelona

At the end of HP Discover Europe in early December I wrote that HP’s internal application of its Moonshot hyperscale technology was impressive, but the presentation left me wondering if this revolutionary technology was finding customers outside HP. Janet Bartleson, group manager for hyperscale portfolio marketing at HP, was kind enough to answer my question by sending a series of short videos of four Moonshot customers ranging from a startup Web services company to PayPal to Purdue University’s HPC center, talking about their experiences. So the answer is Moonshot is gaining traction.

Moonshot is the only hyperscale computing platform generally available in the marketplace, making it a radical departure from normal computing architectures. Hyperscale is based on the philosophy of using inexpensive standardized white box hardware components in massively parallel, virtualized environments where if something breaks it replaced and discarded, not fixed. It was developed by Yahoo!, Google, Amazon, Facebook, and Twitter all of whom build their own hardware.

Massively parallel computing

Moonshot uses consumer grade processors and flash storage designed for mobile systems, which reduces cost and energy, and therefore cooling, demand over standard commercial hardware. HP says it provides up to 89 percent savings in energy consumption, 94 percent savings in floor space, 63 percent lower purchase price, and 97 percent less complexity. John Brown, president and cofounder of serverCondo Hosting, said “When we dove into it and found out what it could do, we decided to take what we were doing and throw it out the door.”

Individually these chips have much less power than the processors in traditional servers. However, in aggregate these massively parallel systems provide enough power to be used in HPC environments such as Purdue University’s computer lab, where it is delivering more compute power for less cost and in a much smaller footprint.

“Moonshot,” says Paul Quick, principal architect at PayPal, “promises to be a real game changer. It is a fundamental new approach to how to how computers can be constructed in distributed architectures.”

One of the most innovative aspects of Moonshot’s design is its cartridge approach to hardware componentization. A single Moonshot 1500 chassis contains 45 cartridges, each an individual plug-and-play server, providing all the connectivity, cabling, etc. Replacing a server is a simple matter of pulling the defective unit out and dropping the replacement in. This is a creative take on the hyperscale approach that makes the don’t fix strategy practical for companies that do not have the huge scale of Google or AWS. It also allows a unit to be sized to fit the need.

Optimizing at cartridge level

A completely software-defined system, each cartridge can be individually optimized for a particular workload through the HP Pathfinder Ecosystem, so different cartridges within a chassis can be used for different workloads.

Its large parallel architecture makes it ideal for many cloud applications including page serving and transaction processing and VDI. HP now runs its worldwide Website on it. Before Moonshot, HP needed 50 Web servers distributed across its six data centers worldwide to handle all the traffic on its site. Now it uses one Moonshot chassis per data center.

“Moonshot,” says Volker Otto, director of HP IT Platform Services, “is a new class of servers, engineered to reduce power consumption and reduce the footprint of servers in the data center.”

Moonshot is an excellent platform for public cloud service providers such as SaaS and VDI-as-a-Service companies and for big data analysis, which often can be done more quickly on a large parallel system than a traditional single CPU system. In these and other applications that are suitable for parallel processing, including many scientific research environments, it can provide high performance with high levels of redundancy at lower cost, smaller footprint, and less power use than standard solutions. And its unique physical architecture allows easy replacement of failing components, simplifying and taking cost out of maintenance.

As a massively parallel system it is not suitable for all workloads, particularly more traditional vertically integrated applications. However, it is a very good choice for cloud and similar systems and essentially can be a “cloud in a box.” This is something that any company that runs its own internal cloud environment or large, heavily trafficked Web environment, or that is considering VDI or other services that require a parallel processing infrastructure, should consider, regardless of whether it is currently an HP customer.

Image and videos courtesy Hewlett-Packard Corp.