The rise of hyper-converged architecture for storage
A comprehensive collection of articles, videos and more, hand-picked by our editors
Confused by all the talk about converged vs. hyper-converged infrastructure in the data center and what it means...
to storage? If so, you are not alone. Consultant Scott Lowe laid out pros and cons of approaches to convergence and how they differ this week during the Modern Infrastructure Summit in Chicago.
Lowe, founder and managing consultant of The 1610 Group, told conference attendees that the rise of convergence is a reaction to storage complexities and scaling issues. Convergence can be a combination of traditional data center hardware products or a marriage of software-defined storage on commodity hardware in a hyper-converged set up.
Convergence can combat storage complexities caused by virtualization, such as the I/O blender and application and storage tiers that have created silos. Lowe said virtualization has made scalability a major issue for storage, and convergence can help solve that.
"Scalability needs to become a feature rather than an event," he said. "Now it is an event, you have to schedule downtime to scale. We want to see scalability baked into everything we buy."
Hardware convergence combines existing products such as a SAN array, switching and compute into one SKU to streamline procurement, deployment and support. It reduces the need for proof of concept testing. A Vblock from EMC's VCE, combining products from EMC, Cisco and VMware, is a prime example. While hardware convergence simplifies buying and set up, it also has a high cost. Vblocks run into "the high six figures or seven figures," Lowe pointed out. And converged hardware does not change the operational process after the equipment is bought and deployed.
Rather than combining traditional storage arrays and networking, hyper-convergence includes storage management software and commodity hardware, often using virtual storage appliances. It enables a VM-centric policy for managing storage, data protection and networking inside one box. But Lowe said hyper-converged is more of a "made-up word" than a technical term, and the definition is open to debate.
"If you have nothing to do on a Friday night, you can go on Twitter and get into a fight about what hyper-converged is," he said. "My baseline definition is that it's storage and compute in one appliance with workloads that run side by side with a VSA [virtual storage appliance]."
Put another way, hyper-convergence means "no more SAN. We're back to the days of direct-attached storage. Virtual machines manage storage that's local on their nodes."
Hyper-convergence can be bought as software only from vendors such as VMware (Virtual SAN) and Maxta, or as an appliance with software installed from vendors such as Nutanix, SimpliVity, Scale Computing, and VMware's EVO: RAIL partners.
He said the cost of entry is reasonable, "five-figure reasonable in some cases," with systems ranging from SMB to enterprise. Customers scale by plugging another appliance into a cluster.
Lowe said hyper-convergence streamlines operational efficiency and suggested a hyper-converged system when setting a central policy is important and cost is a key factor. Customers can replicate data and set VM policies that follow the VMs as they move around the hyper-converged cluster.
Hyper-convergence is a good fit for use cases such as virtual desktop infrastructure, remote offices and primary storage for small companies, but not for all storage needs.
One audience member asked Lowe about scaling limitations with hyper-converged systems because you cannot scale capacity or compute independently. Lowe agreed that could be a problem but said "the operations improvement can offset that. Hyper-convergence is not necessarily a silver bullet for every use case. It's another arrow in your quiver."
Analysts predict 2015 will be a hot year for hyper-convergence
Defining the differences of converged vs hyper-converged infrastructure
How hyper-convergence reduces strain on servers
The difference between converged and hyper-converged lies in the software