HCI Doesn't Converge

Most Hyperconverged infrastructure (HCI) vendors claim scalability as one of their key advantages. The phrase "just add a node" is a common mantra of HCI vendors, but because they lack the proper integration of the data center stack, "just adding a node" comes with limitations.

Scaling the Network

The first problem will be networking the new node into the existing cluster and infrastructure. Since most HCI solutions provide little networking capabilities beyond the virtual switch, which they borrow from the hypervisor, networking complexity increases as you add each node.?

Scaling Processing Power

Most HCI can't aggregate processing power. They can add processors with each node, but if you have an application that needs more than the processing power of a particular node, you can't chain processors together. You are limited to the processing power of a node.?

Furthermore, most HCI solutions can't mix processors within the infrastructure. You must set up a separate HCI instance if you started with Intel CPUs and want to bring on AMD CPUs. You bought HCI to eliminate silos, and legacy solutions force you to add them! If you need or want to use a combination of Intel, AMD and Nvidia, you'll need create silos of each type.

Scaling Storage Capacity

When it comes to storage, as you add nodes, most HCI solutions can aggregate storage capacity. That said, as capacity expands, the HCI storage software struggles to keep up and eventually hits a limit. Most HCI solutions can't handle more than a couple of petabytes of capacity before adversely impacting performance.

A few HCI solutions have developed "capacity nodes" so that you can mix nodes of different storage profiles. The problem is that mixing these nodes introduces additional compromises. We've seen vendors suggest turning off deduplication or changing your drive-failure protection technique if you mix performance storage nodes with capacity storage nodes. Ultimately, the not-so-subtle recommendation is to create another cluster of just capacity nodes. If you're keeping count, that's three silos your converged infrastructure is forcing on you.?

Scaling Storage Performance

Storage performance presents another scaling challenge for HCI. For example, if you add NVMe flash drives, the HCI cluster does not keep pace with the full potential performance of the drives themselves. There are numerous reasons for the performance gap:?

• The storage software runs as a virtual machine (VM), suffering from the same virtualization tax as any other application.?
• The storage VM features create overhead. High-performance drives will expose the vendor's often clumsy implementation of deduplication and drive failure protection. ?
• The HCI's network inefficiency. Most HCI solutions have little control over the network and can't optimize for the massive east-west traffic standard in scale-out architectures using NVMe or faster drive technology.

Silos of Convergence

Beyond not scaling out, most HCI architectures can't scale small to support small data centers and Edge Computing environments. You need separate infrastructure software for your Edge locations and your enterprise. That's four silos.?

They also can't mix nodes of different types to various workloads within the same infrastructure and deliver overall infrastructure longevity. That now brings the number of silos HCI needs to more than what the data center had with three-tier architectures. It is why three-tier architectures significantly out-pace HCI in terms of sales.?

Convergence Needs Ultraconvergence

Hyperconvergence failed to deliver because it didn't eliminate the complexities of dealing with the three-tier data center stack. With HCI, virtualization, storage, and networking all run on the same hardware but are still separate layers. Ultraconverged Infrastructure (UCI) rotates the stack, creating a single, linear plane and eliminating those separate tiers., combining them into a data center operating system with a common code base.?

UCI solutions like VergeOS can start at one or two nodes for Edge Computing, scale to three or for nodes for small data centers and hundreds of nodes of different types for enterprises. Each time you add a node, you experience a linear increase in processing power, storage capacity, and performance. VergeOS's efficiency gives you more performance and capacity to support more workloads on less hardware than any HCI solution. It also provides complete L2 and L3 networking functionality, making "just add a node" as easy as possible.?

VergeOS is convergence. Its efficiency drives down IT spend and complexity by as much as 70%.

To learn more, follow us on LinkedIn.?