As service providers flock to virtualization and other New IP technologies, they are demanding more and more from their virtualized environments. This has, at times, been troublesome due to performance and latency issues which have raised questions over the long-term survival of the species known as the vSwitch.
"Performance and latency have notably been pain points when deploying virtual machines on a vSwitch for software-based network virtual functions such as a vRouter or a vFirewall," Jason Nolet, senior vice president of Brocade's Data Center and Enterprise Networking Group, told The New IP in an email interview.
Roz Roseboro, senior analyst, Heavy Reading, also takes note of these issues, telling The New IP via email that she sees them as indicative of a "need to re-architect the VNFs to account for different underlying architectures."
Still, that doesn't change the fact of demand. "Customers want a high density of VNFs on a single server," according to Nolet. "VSwitches sometimes prove to be performance [and] scale bottlenecks in [these] scenarios."
Consequently, Nolet predicts: "The vSwitch itself will likely undergo transformation over the next several years, including the possibility of its disappearance entirely."
The possibility of vSwitch extinction is a bold outlook, indeed, but Nolet sees great potential for vSwitch transformation in the current container revolution.
"Recently the introduction of Linux containers has generated some interest as alternatives to hypervisors because cost and management overhead was a pain point in VMware deployments," reports Nolet. "But availability of management tools will ultimately dictate broad-based usage."
For all of their benefits, however, not everyone agrees that the container stands to replace the humble hypervisor outright.
"Containers simply do not provide the same flexibility that hypervisors provide within a virtual hardware environment," Mark Coggin, senior marketing director for Red Hat's Platforms Business Unit, told The New IP in a separate interview.
More importantly, for Coggin, containers and hypervisors are not an either/or proposition. "With no guest operating system lighting up virtual hardware, applications have to be native to the container host. It's a tradeoff between flexibility and efficiency and ultimately why we will see containers and hypervisors often used in complementary, not competing, roles."
Even so, Nolet points to additional lines of transformation along which the hypervisor is proceeding. While Nolet concedes that tools like Intel's Data Path Development Kit have helped to address performance and latency issues with virtual functions, both he and Roseboro believe that vSwitch data path acceleration can help even more.
"In the future," Nolet says, "actual hardware offload of the data path on to server NICs [network interface controllers] is seen as a means to achieve even higher performance."
Similarly, Nolet points to vSwitch bypass mechanisms, such as Single-Root Input/Output Virtualization ("SR-IOV"), as another way of letting the physical NIC do more of the work through direct access.
"This is generally considered when low latency is extremely important for the application," says Nolet.
For now, however, Nolet's visions are but pipe dreams for the more integrated, efficient enterprise.
"Today SR-IOV is primarily used when VMs that need to communicate are resident in the same server and uses the server NIC as their low latency connectivity switch. Tools are lacking for wider usage of SR-IOV," Nolet laments. He further notes that there is also a lack of tools to manage hardware-accelerated environments.
"Both service providers and enterprises will be careful in picking options," Nolet says. "Tradeoffs will primarily be between the need for performance and complexity of management."
Coggin, however, sees a bright future for virtualization on all fronts thanks to open source technologies. Daylight-bright. "The [future vSwitch] will look a lot like the technologies being developed by the OpenVSwitch and OpenDaylight Foundations -- open-source solutions built on common standards that provide a truly open networking stack and management layer," says Coggin.
Coggin is emphatic on one point, however; the hypervisor tradition will carry on -- not to be replaced.
"Virtualization, and the hypervisors supporting it, is … already entrenched in the enterprise datacenter. It's been around for ten-plus years and runs many of the most critical business applications in the enterprise world while remaining a relatively cost-effective solution for monolithic, traditional applications through server consolidation," says Coggin. "Despite all of the doom and gloom surrounding the hypervisor, it, and traditional virtualization, aren't going the way of the dodo."
— Joe Stanganelli,
Freelance Contributor, special to The New IP