Cluster Resources, the middleware vendor that brought its Moab cluster management technology to market, is making a major move to broaden its footprint beyond the company’s high performance computing base. As of this week, the company will be known as Adaptive Computing to reflect its expansion into commercial datacenters and private cloud environments. The new organization will encompass two business units: Cluster Resources, for its traditional HPC customers, and Adaptive Computing for the commercial enterprise side.
In fact, the company had been heading in this direction for some time. Founded in 2001, Cluster Resources developed its Moab cluster workload management product line around the Maui open source technology. At the time, cluster computing was generally confined to high performance computing. But over the next several years, clusters thoroughly infiltrated the enterprise, and by 2007, Cluster Resources’ business was split 50-50 between the HPC and the commercial datacenter markets. Commercial customers today include Yahoo and a number of financial institutions, among others.
As clusters became the platform of choice throughout the industry, systems grew larger and were employed to support a much wider variety of workloads — database machines, real-time transactional systems, Web application platforms, and so on. Today the trend is to consolidate datacenter infrastructure everywhere, and that means these large multi-faceted facilities now resemble supercomputers to a great degree. In both environments, datacenter-level virtualization and workload automation are often the norm.
Datacenter-level virtualization and automation also happen to be the foundation for cloud architectures. In a cloud, all of the infrastructure (compute, storage and networks) as well as software licenses, are treated as a shared pool of resources. So the same approach of abstracting software from individual hardware components is now being applied across the entire datacenter landscape.
Since Moab’s main strength is managing disparate architectures at the level of the datacenter, Cluster Resources saw this convergence as an opportunity to leverage its core technology. With that in mind, the company has introduced the Moab Adaptive Computing Suite, the flagship product of the datacenter business unit. In addition to the workload management tasks, it also includes features that support the type of computing more commonly associated with the typical enterprise application, namely system-level virtualization. Unlike HPC, in other large-scale computing environments, it’s common to have multiple applications running concurrently on a cluster node, or even a single CPU.
“The main difference that we’re seeing between supercomputers and commercial enterprise sites is that the workload types are slightly different,” explains Peter ffoulkes, Adaptive Computing’s vice president of marketing. Specifically, there tend to be a lot of transactional workloads in the enterprise. But ffoulkes also notes there is a blurring of workload characteristics between the two areas. For example, enterprise applications such as business intelligence (BI) are both data and resource intensive. In fact almost all informatics applications have these profiles and they span HPC and the more traditional enterprise space.
The big OEMs are also reflecting this convergence in their latest servers aimed at scaled-out datacenters and in their general approach to next-generation computing. IBM (Dynamic Infrastructure), HP (Adaptive Infrastructure), and Cisco (Unified Computing) are all pushing their system architectures into this model, in one variation or another. Since IBM and HP are also close partners with Cluster Resources, we can expect to see the new Adaptive Computing technology show up on future deployments.
According to ffoulkes, their relationship with both IBM and HP has been extended. For IBM, this means that Moab-based products are now being sold with IBM part numbers alongside Big Blue servers (for example iDataPlex systems) and the xCat and Tivoli provisioning products. HP has also expanded the partnership to integrate the Moab technology into its HP iLO (integrated lights out management) and HP SA (server automation) software to create a “Dynamic Workload Utility” for scaled-out environments.
Although the HPC side of the company is now under its own business unit, ffoulkes said they will maintain their commitment to the supercomputing market. Today 60 percent of the top systems are powered by Moab technology, including the top two machines: Roadrunner at Los Alamos National Lab, and the Jaguar system at Oak Ridge. More recently, the University of Southampton announced it had ordered a 1,000-node IBM (iDataPlex) supercomputer that will include the Adaptive HPC Suite for workload management. That system is intended to run both Linux and Windows applications in a wide range of research areas, including climate, pharmaceuticals, bioscience, nanoscience, medical and chemical systems, transport, the environment, and engineering.
One fortuitous side-effect of the company’s dual focus is that organizations that run mixed HPC-enterprise workloads can use Adaptive Computing as a one-stop shop for workload management. For example, a bank may run transactional workloads during the day and risk management, portfolio pricing and the Sarbanes-Oxley compliance reporting overnight. If possible, the firm would like to use the same infrastructure to do all this. While not that common, ffoulkes says some larger organizations are motivated to build these scaled-out generic datacenters that can be repurposed for heterogeneous applications on-demand.
“We’re very much seeing this type of convergence, but it’s happening in a rather a spotty sort of fashion,” says ffoulkes. “It really depends upon where companies are starting from and what they’re able to do. But we’re seeing it on both sides — in the HPC world and the commercial datacenter world.”