Issue link: https://cp.revolio.com/i/661564
THE DATA CENTER JOURNAL | 5 www.datacenterjournal.com n ew industry trends can easily sweep up the unwary into untested strategies and technologies that end up costing far more than they promise to save. Aer an initial wave of enthusiasm oen comes a pullback as both vendors and customers try to find a niche for the technology while jettisoning the expectation that it will solve all problems. e data center industry is no exception to this process—something akin to what re- search firm Gartner describes as the "Hype Cycle." e key, in any event, is to evaluate the information as objectively and sensibly as possible before adopting a technology, regardless of whether it's brand new or an industry staple. at goes for hyperconver- gence as well. What's in a name? "Hyperconvergence is a term used to describe soware-defined-computing solutions that are purchased in a building- block fashion and that through orchestra- tion allow major resources to appear as services available everywhere," said Mark Harris, VP of Marketing for Pluribus Networks. It sounds a bit like a grand uni- fication of data center IT, and in a sense, that's what it is. Harris added, "e solu- tion includes all of the relevant comput- ing, storage and networking components, which are all soware-defined and can be configured on demand to provide the nec- essary data center–wide resources for new applications to be deployed quickly." One goal of hyperconvergence is reduce complexity. Why go buy all the individual parts you need to build a car, for instance, when you can buy one that's fully manufactured and ready to drive? According to Deep Bhattacharjee, Head of Product Management for ZeroStack, "e idea behind hyperconvergence is simplification of the hardware/soware ar- chitecture, so that customers don't have to buy storage from one place and compute from someplace else, then put these things together using complex integration steps." e unit of a hyperconvergence solution is not therefore a storage device or a networking device or a compute server—it's a combination of all three in an appliance that's run by a hypervisor, which unifies the different parts into a whole. is modular approach is designed to increase scalability and reduce the effort of the data center operator: effectively, its aim is to deliver a ready-to-drive car rather than just the wheels, engine block, transmission, frame and so on. "e hyperconverged data center has fewer 'moving parts,'" said Harris. "In most cases each rack of equipment can be treated as a single management entity, rather than 25 or more individual devices that need to be managed and maintained." e hypervisor—the soware "glue" that governs the resources—turns the data center into a facility with resources on tap. Instead of offering individual bottles each containing some fixed amount of the re- source, it's a keg. Harris added that "as far as soware goes, although the applications doing work can be far and wide in their function, the platform itself—the essential computing, storage, and networking—is all soware defined. ey can be viewed as a pool of resources all orchestrated by the platform itself." He also describes the ap- proach as delivering data center resources in the form of services to each "customer," whether it be a VM, an application or whatever the designated user. is venDor LoCk-in a LoCk? e first dilemma that may come to mind when contemplating the hypercon- verged data center is vendor lock-in. If you buy a car from a dealer, for instance, you generally can't expect to get a Ford engine, GM transmission and Honda body. e same concern applies to hyperconverged infrastructure, although efforts to promote open-source technology aim to ameliorate it. Data center operators are interested in technologies that ease their task, but not always at the expense of flexibility. A ready-made car might be the easiest option, but it probably won't deliver the kind of specialized performance that, say, a stock car driver is looking for. Vendors, however, have at least a surface incentive to limit customers' ability to use compet- ing products. Harris notes, "Although many of the current hyperconverged solutions are vendor specific, there are many initia- tives underway that use the work of open source. Projects such as OpenStack and Apache's Mesos enable resources to be virtualized and consumed, and as the data center industry becomes more insistent on open computing, we will likely see hyper- converged providers follow suit to exist in an open ecosystem." Some vendors argue that hypercon- verged infrastructure is flexible by nature and therefore not subject (or just less subject) to lock-in. Regardless of whether that's true, a likely exception is in the case of vendor-supplied alternatives to standard hypervisors (whether proprietary or open). But such solutions don't represent the entire market. Counting the Cost One can rightly expect a fully manufactured automobile, for example, to cost more than the sum of its parts. Such an expectation, at least on the capital side, therefore seems reasonable with regard to hyperconverged infrastructure. But that doesn't mean the additional cost isn't worth it: how many people have the time and expertise to build their own cars from scratch? Such capability requires lots of previous experience climbing the learning curve, not to mention a few expensive tools. "e costs of hyperconverged solu- tions today tend to be slightly higher than their discrete counterparts for the simple reason of convenience," said Harris. But he also notes the fact that the price tag isn't everything. "Simplistically, the packaging and integration of many granular compo- nents into a final modular solution has a cost to the supplier but an immense value