At the Gartner Data Center conference in December 2016, analyst George Weiss gave a presentation on the “Impact of Compose-able on Next Generation Hyper convergence”. In that presentation, he claimed that by 2018, elements of compose-able functionality will ship in 20% of converged and integrated systems. I have an issue with this premise because it assumes that only converged (or hyper converged) or integrated systems have elements of composability. At DriveScale, we believe that any server design can be an element of a compose-able infrastructure.
At its heart, composability simply means that individual parts of systems can be combined with parts of other systems to form a newly composed product/solution that is capable of performing general purpose functions. The end result is that you can pool your resources together and then allocate parts or the whole of it for specific workloads. Dis-aggregation in itself is not enough to create a workable solution. You need a tailored management system that can re-assemble parts to make functioning systems that are optimized for each type of workload. A few LEGO blocks that can be assembled on demand to make different things for different uses.
All the benefits that George presents in his talk apply to the DriveScale solution. These include: the ability to grow or shrink cluster node counts for just-enough compute; being able to expand or reduce storage to meet processing needs; reassigning compute nodes and drives when failures occur; responding to workload shifts; and, creating a public cloud like elasticity by managing compute, storage and the fabric individually. The key difference is that DriveScale achieves all this with commodity hardware instead of proprietary system designs that result in tremendous cost.
Finally, a significantly important factor is the scale of composability. If all you get is composability at the scale of a few servers in a blade chassis, or at the rack level, then you cannot accrue the benefits of a truly compose-able infrastructure at scale. What you need is composability at the data center scale. That is to say, all the components of the data center should have the ability to be assembled on demand.
All the OPEX benefits that George states in his presentation accrue when customers use DriveScale in their Hadoop environments, like: Better resource utilization; lower upgrade costs; intelligent workload placement; reduced operational effort, and; reduced provisioning time. So, what’s to lose when you pick DriveScale? NOTHING.