SAN FRANCISCO — The next phase of composable infrastructure, according to Liqid CEO and co-founder Sumit Puri, involves 5G networks as a fabric type.
SDxCentral caught up with Puri at VMworld, where his company showcased VMware’s hyperconverged infrastructure (HCI) software running on top of Liqid’s disaggregated composable infrastructure. We talked about how this represents the “next evolution of hyperconvergence and virtualization” (more on that in a bit).
But first: composable 5G?
Puri was quick to add that this is still a few years away. “Kind of like blow your mind in the future,” he said. But it’s something that Liqid is already working on with networking equipment providers using Liqid’s software.
“We are working with companies where we can build adapter cards that will put that pool of resources onto the 5G network, and then our software will grab whatever is on the 5G network and compose them into servers,” he explained. “We enable physical resources to be shared in multi-tenant environments over wireless infrastructure.”
Potential customers include communication services providers and managed services providers that own multiple data centers, 5G edge providers, and large enterprise customers that operate private clouds, Puri said.
A couple future use cases he envisions include allowing customers to share physical devices across multiple data centers over wireless 5G and compose a GPU from the data center to a notebook in the field.
“It’s still a few years away,” Puri said. “But with the bandwidth and latency that 5G can provide, we can envision an environment where 5G is a wireless fabric type that can be leveraged for disaggregated infrastructure.”
Liqid Composable Infra for HCI
In the meantime, however, Liqid is making strides in what Puri says is the next step in HCI: virtualization software deployed on Liqid’s composable, disaggregated infrastructure. This addresses what he says is the fundamental problem with hyperconvergence: “it’s converged.”
At VMworld the company showed how VMware’s vSAN-based hyperconverged infrastructure can orchestrate resource allocation via Liqid APIs. This allows companies to maximize their existing hardware and ultimately save money on infrastructure costs.
“Our belief is the next evolution of things like hyperconvergence and virtualization is not deploying those on static, converged boxes,” he said. “It’s deploying that same software layer on top of disaggregated pools of resources, and then the hypervisor layer can say give me more storage so I can chop it up.”
This creates bare-metal servers through software, which effectively treats bare metal hardware as disaggregated resource pools. Traditional HCI runs storage on a hypervisor, but these systems don’t typically allow independent scaling of compute and storage resources. So if a customer needs more resources, he buys more servers — and more software licenses tied to physical CPUs.
“With disaggregation we don’t do that,” Puri said. “When my x86 processor runs out of resources to virtualize, I recompose, bring in more resources, and then virtualize on top of them. I’m no longer confined to the model of fixed boxes to software license ratios. We can break those ratios.”
Add-On Accelerator Tech
This also allows customers to add accelerator technologies like GPU, FPGA, and Intel Optane memory on demand and utilized as shared resources in quantities that best correspond with a given virtualized compute task.
And it saves money. “Most of our TCO is 60% to 70% percent savings you can achieve through disaggregation,” Puri said.
While Liqid started with VMware, “because we see that as the biggest opportunity,” it’s talking with other HCI providers including Nutanix and Pivot3, Puri said. “We think that eventually all of those guys would benefit from changing their model from stacking boxes to disaggregating boxes.”