Is InfiniBand about to make a dramatic comeback in the enterprise? Quite possibly, now that the cloud is placing greater emphasis on network agility and data fluidity in high bandwidth network infrastructures.
InfiniBand, if you recall, was originally pitched as a general enterprise solution, although it quickly found a more receptive audience in high-power computing (HPC) environments once lower-cost Ethernet solutions proved to be good enough for most enterprise applications. In the cloud, however, the ability to move data from constantly shifting, mostly virtual, endpoints is of tremendous value in that it gives both enterprises and hosted service providers a leg up when it comes to meeting desired service levels.
The Taneja Group highlighted a number of reasons why InfiniBand is likely to excel in the cloud. For one, it provides the basis for a high-speed fabric that can optimize web-scale application performance, scale-out storage architectures and virtual I/O environments for mission critical functions. With CPU performance and both core and socket counts on the rise to accommodate the demands of the cloud and big data, a robust interconnect is the only way to ensure those gains are not lost on the way to the user.
A number of leading cloud providers are already turning to InfiniBand as a means to gain a competitive edge. German IaaS provider ProfitBricks recently tapped Mellanox for end-to-end InfiniBand infrastructure for its U.S. and European data centers. The company says it can increase transfer rates by a factor of four and reduce latency 10-fold compared to existing 10 GbE platforms using dual-port InfiniBand server connectivity. At the same time, the company can introduce rapid scalability without sacrificing network efficiency or service availability.
Mellanox is also pursuing a number of channel opportunities as it seeks to leverage the cloud-InfiniBand relationship. The company recently inked a deal with systems developer Quanta QTC to devise cloud-ready server and storage solutions using InfiniBand-based fabric technology. The deal is most likely to center around the ConnectX-3 portfolio as a means to increase bandwidth and offload network transport responsibilities from the CPU so they can devote more resources to application performance.
The cloud angle also adds new perspective to Intel’s purchase of QLogic’s InfiniBand line earlier this year. At the time, most people saw it as a bid for even greater HPC prowess, which is a still a valid goal. However, the cloud adds a much broader commercial opportunity for InfiniBand in that enterprises of all stripes will be more likely to shell out extra dollars for a high-speed fabric that can help them lower costs and drive greater efficiencies in their data infrastructure. It just might be the engine the fuels Intel’s dreams of exascale computing in which clustered resources are propelled to truly massive proportions.
About the only thing wrong with InfiniBand’s plans for the cloud is that that enterprise industry has already thrown much of its infrastructure to Ethernet. That’s quite an installed base to simply cast aside, unless there turns out to be a compelling reason to do so.
In that light, then, InfiniBand backers will not only need to provide a robust fabric, but the must-have application sets that can take advantage of it.