Dell Wants To Build The Network Hypervisor

Dell believes the future of networking will mirror that of compute: software defined, virtualized, and based on standards.
A couple of weeks ago I spoke with Adnan Bhutta, Dell's vision for the future of networking involves a de-coupling of physical hardware from software, and that cloud attitudes are driving much of the change.

“I think the networking industry is going through the same things that the compute industry went through fifteen years ago,” says Bhutta. “People had these mainframe models. Essentially you had Dell's vision of the future, networking will go the same way, decoupling the server hardware from the software that runs on it.

Until recently, your choice of networking operating system has been tightly coupled to your choice of hardware: Cisco switches ran IOS, Juniper switches ran Junos, etc. “Traditionally, like a year and a half ago, you had to buy our switch, our software, everything from us,” says Bhutta. Now you can buy switch hardware from Dell's switching OS, or an operating system from Cumulus, BigSwitch, IP Infusion, or Pluribus.

Bhutta also highlighted Intel or AMD chip implementation from the operating systems that run on them.

The use of such a standard would make it significantly easier for network software vendors such as Intel, and Mellanox can concentrate on making ASICs that forward packets faster. Not incidentally, it would also help to commoditize the switch port business which would help these vendors at the expense of Cisco.

“The bottom line to all of this is ease of management and deployment,” says Bhutta. “Software defined networking's main value proposition is that it has to save you OpEx.” He cites a comparison between the number of people required to administer servers now compared to what it used to be, and then highlights the disparity with network administration. “If you look at the number of server admins you have per server versus the number of network admins you have per switch, the numbers are mind-boggling!” he says. “You have a lot fewer people to manage a lot more servers than you have on the networking side.”

Bhutta is referring to the highly virtualized x86 server world of today, while networks more closely resemble the pre-virtualization days. VLAN may stand for Virtual Local Area Network, but the devices are more akin to a single server running many applications rather than a physical device running multiple instances of a networking operating system. If the equivalent of x86 for networking were to become widespread, then imagine what the equivalent of VMware for networking might look like.

Image a piece of switch hardware that, like its x86 server equivalent, could run multiple switch/router operating systems simultaneously. You can sort-of do this already by using virtual machines in an x86 based hypervisor environment, but x86 chips are designed for general computing workloads, not the highly specialized world of moving packets around. Switching silicon is several orders of magnitude faster at this specific task, just as GPUs on graphics cards are much better at floating point math than general purpose CPUs. Keep in mind that virtualization was substantially helped by the addition of specialized hardware features in x86 CPUs: VT-x for Intel, and AMD-V for AMD.

Switch hardware that can run multiple operating systems would be like using the same x86 server model to run Windows or Linux depending on what you install on it. But in this modern, hypervisor dominated world, that server could run also Xen or ESXi, which add a lot of flexibility for the relatively small price of the hypervisor overhead.

That's the kind of world that VMware of this switch hypervisor world?

This article first appeared in Forbes.com here.