Cutting edge: IT's guide to edge data centers
A comprehensive collection of articles, videos and more, hand-picked by our editors
You know a tech has gone mainstream when its product category starts to split along four lines. That's the case with hyper-convergence and is evidence that it's gaining strength and acceptance. The hyper-converged market represents approximately $1B in sales and is expected to grow in excess of 50% compound annual growth rate over the next three years.
Hyper-convergence emerged as an alternative to building an IT infrastructure in layers. Instead of buying compute, storage and other infrastructure elements separately, vendors like Nutanix and SimpliVity tightly integrated them in a scale-out architecture. Compared to converged products, such as HPE ConvergedSystem, NetApp FlexPod and VCE, deployment and management was much easier, Capex was lower by a factor of four and scaling was trivial.
But hyper-convergence is about more than tight integration of compute and storage. It's about support of multiple hypervisors, virtualization operations, true application mobility across hypervisors, capacity planning across the infrastructure, integrated end-to-end data center management and so on. Cloud integration is becoming another distinguishing feature, with an ultimate vision of replacing entire infrastructures with a new web-scale hyper-converged architecture.
The hyper-converged market is splitting along four lines:
- Primary storage
- Primary plus secondary storage
- Secondary storage only
- Edge applications and storage
Hyper-convergence primary storage
This is the best-understood category as it's the essence of the original theme: A scale-out architecture for primary storage applications where the boundary between compute and storage is blurred and both are combined in a single node. For all products in this category, the focus shifts from managing storage to managing a VM -- everything becomes VM-centric.
Hyper-convergence of primary and secondary storage
When hyper-convergence started taking off, we saw one vendor -- SimpliVity -- start differentiating itself along the secondary storage dimension. They presented their product not only to solve the primary storage problem but also for many secondary storage use cases. Data protection, replication, WAN optimization, deduplication, compression and other functions normally associated with secondary data handling were all part of SimpliVity's story.
Of course, most other vendors have at least some of these functionalities built into their products. But SimpliVity made a clear architectural distinction: by deduplicating and compressing data inline at the time of creation or ingestion, they would integrate these secondary uses of data so tightly that one wouldn't have to buy separate products for any of the functions. It demonstrated complete hyper-convergence of primary and secondary storage use cases.
To be sure, Nutanix too provides snapshot-based backup/restore and replication, but prefers to work with third-party data protection vendors to solve large-scale secondary storage issues. They place a much stronger emphasis on other aspects of hyper-convergence mentioned above, including Web-like scalability and support of their own hypervisor, Acropolis.
Today, you can buy a product in the hyper-converged market focused on primary storage and separately buy individual products for secondary storage, or buy one that has it all tightly integrated.
Hyper-convergence of secondary storage
This hyper-convergence segment was created on the premise that what's good for primary storage is also good for secondary storage. Clearly, such a product would make a solid alternative to primary storage hyper-convergence as well as combined primary/secondary hyper-convergence, and would cover a wide spectrum of infrastructural functionality.
Noting that secondary storage functionality today is far more than backup/restore, replication, data deduplication, compression and WAN optimization, new vendors such as Cohesity and Rubrik embraced those and other functionalities such as DevOps, copy management, big data analytics and more to effectively create a new category.
While the two vendors look similar on paper, their architectures and strategies differ quite a bit. Cohesity, for instance, offers (or plans to) all that functionality but will still work with third-party products via open APIs; Rubrik, on the other hand, will strictly offer their own functionality. But they share a premise: tightly converge as many of the secondary use cases as possible and work cooperatively with other hyper-converged market offerings.
"Edge" is the latest variation, with only one purveyor today: Riverbed. The company focuses on ROBO sites, where even hyper-converged systems are too much to manage with little or no IT expertise available. Their simpler configuration for ROBO is essentially IT-free but offers all the benefits of local application performance and convenience.
Riverbed pioneered WAN optimization and consolidation with its SteelHead technology. The original idea was to remove all IT infrastructure in the ROBO and serve the application entirely from the data center. Data would reside 100% in the data center and the best of all data protection methods, including DR, would be applied to it. With a SteelHead appliance at the edge of the data center and one in each ROBO, the appliances would use WAN optimization to mitigate the effects of long distance latency. The result would be "local-like" application performance for ROBO users with only user devices at the remote sites. One would get excellent application performance with no IT infrastructure to manage.
But not all applications fit nicely into this category. WANs are notorious for being flaky, and may go up and down a few times a day. For some applications, work cannot be stopped, even for short intervals. Some applications simply couldn't be consolidated in the data center; they needed to run locally in the branch offices. If the exceptions called for IT infrastructure in the ROBO, and required data to be stored onsite in the branch office, then we were back to square one with IT infrastructure to manage in each ROBO.
Riverbed recognized these challenges and developed SteelFusion, an appliance that would sit in the data center and at each ROBO, and the application, in the form of a VM, would be "projected" to the branch offices to run locally on the appliance with true local performance. But the master application and its data (VM) would be located in the data center where it could be managed and protected.
The ROBO appliance -- SteelFusion Edge -- is the SteelHead WAN optimization integrated with virtualization and storage cache technologies. Any new data created or changed at the branch office would be instantly applied to the master VM, using all the principles of latency mitigation and WAN optimization built into SteelFusion. Most importantly, the application would continue to run even if the WAN failed, and performance would be truly local. If the branch office disappeared for some reason, the applications could be fired up elsewhere using another SteelFusion Edge appliance. The application and IT management would stay in the data center with the branch office remaining "IT free." The data center could operate on separate compute, storage and networking, or be converged (as in VCE or FlexPod) or be hyper-converged, but the branch office would use Riverbed's SteelFusion Edge. A function of the product called FusionSync automatically fails over all ROBO operations to a secondary data center in case of the loss of the primary data center. A one-click failback is used to return operations to the primary data center when it becomes operational again.
The hyper-converged market is no longer a monolithic category. It's forming tributaries and subtle variations are cropping up. I consider this ratification that this style of computing has serious merit as demonstrated by the many users who are already enjoying its benefits. But no two IT shops are alike, so you'll have to decide which variation of hyper-convergence is right for your company.
Should you enter the hyper-convergence market?
Make sense of the crowded hyper-converged marketplace
Hyper-converged products move from hype to reality