Thanks to the ubiquity of the X86 server platform, the Kubernetes container controller, and the KVM server virtualization hypervisor, it is relatively easy to compute like a hyperscaler or cloud builder. But networking like one has been another matter entirely. But it is getting easier.
In July 2015, Google gave the world Kubernetes, a sort of foundation based on its international Borg and Omega container and virtual machine controllers, for its own enlightened self-interest reasons: Having the world speak your metaphor is almost as powerful as having the world adopt your software. Kubernetes immediately became the de facto container platform, consigning Mesos, Docker, OpenStack, and a number of alternatives (including the original OpenShift from Red Hat, which was based on a completely different container metaphor and set of code to express it) to the dustbin of history.
Microsoft grabbed the Linux kernel and created the Software for Open Networking in Cloud (SONiC) network operating system and its related Switch Abstraction Interface (SAI) for its own hyperscaler and cloud purposes, and donated it to the Open Compute Project in September 2015 for much the same reason that Google let go of Kubernetes: To foster a truly open layer of software in the datacenter. In the ensuing years, Dell, Hewlett Packard Enterprise, Big Switch Networks (now part of switch maker Arista Networks) and Mellanox Technologies (now part of Nvidia) open sourced their network operating systems in the hope of creating the analog of Linux computing on networking devices. But it was pretty apparent even three years ago that SONiC was winning the war of the NOSes, with all of these efforts, as well as that of Cumulus Networks (another open source NOS supplier that is also part of Nvidia) failed to go mainstream in the datacenter for switching.
The mystery to us was that Red Hat (now part of IBM) or Microsoft did not see the opportunity to create a commercially supported distribution of SONiC. Dell and Apstra did back in 2020, but with limited switch coverage. A startup named Hedgehog, founded by a bunch of former Cisco Systems executives, made a bit of noise last October when it announced its own commercial-grade SONiC distribution. And now Aviz Networks is entering the picture with its Open Networking Enterprise Suite, or ONES.
But rather than creating a SONiC distribution, Aviz Networks has created a network management tool that makes it easier for companies to adopt and manage community SONiC code alongside of the distributions from Dell, Apstra, Hedgehog, and Broadcom (which quietly has its own SONiC distribution). Importantly, the ONES tool can also pull telemetry out of other network operating systems such as NX-OS from Cisco, EOS from Arista Networks, and Cumulus from Nvidia to help manage traffic across many different switch and NOS combinations.
This is an important tool because, unlike Microsoft, which went SONiC top to bottom in the Azure Network underpinning its Azure cloud three years ago, most service providers and large enterprises that want to go SONiC in the network will nonetheless have a mix of hardware and software as they transition to whitebox switches running SONiC.
“Our mission for the next year is enabling large enterprises to replicate the success of hyperscalers with SONiC,” Vishal Shukla, co-founder and chief executive officer of Aviz Networks, tells The Next Platform. “Whatever Microsoft has done, companies like eBay, Cloudflare, Wal-Mart, and Target should be able to do without investing in an army of engineers. And it is important that we are the only neutral entity in the SONiC ecosystem. We don’t make a switch ASIC. We don’t make a switch. And we don’t make a SONiC distribution. We have a 35 person team of NOS experts but we don’t have a NOS of our own.”
So maybe a better metaphor is that Aviz Networks wants to be the Datadog of networking? Importantly, the mission is broader than that, with Shukla hinting that ONES will eventually be augmented with AI algorithms to help manage traffic across SONiC networks, and maybe even non-SONiC networks.
Aviz Networks was founded in 2019 and has been self-funded until it raised $4 million in seed funding in April last year. Shukla says that rather than take a top down approach and create a SONiC distribution and try to push it into enterprises who want to have the benefits of an open NOS, Aviz Networks has taken a more patient approach and listened to how Microsoft, Alibaba, and others are using SONiC and has figured out how to help organizations support SONiC in production. And by staying away from the venture capital money and trying to create a Red Hat for Linux-based networking, Shukla says Aviz Networks can grow at the speed of SONiC adoption instead of trying to force SONiC adoption at a faster rate than the market is ready to convert to it.
Not that the market is not growing. Gartner reckons that by 2025, a mere two years away, 40 percent of large enterprises that have more than 200 switches in their infrastructure will be running SONiC in production.
Aviz Networks started out by automating test scripts for the various SONiC implementations, and eventually rolled this up as the Fabric Test Automation Suite, which is bundled up in a virtual machine and which does continuous validation for both datacenter and edge implementations of SONiC. At this point, says Shukla, almost all of the SONiC players – those creating distros, or ASICs, or switches – are using FTAS to validate their platforms. The other thing that the Aviz Networks team created as one of the companions to ONES is called the Open Packet Broker, which is itself based on SONiC and which captures, aggregates, and analyzes network traffic and feeds data into network monitoring and operations tools that run the network.
The ONES stack has four components. It includes an Orchestration module that is essentially a fabric manager that lets network administrators compose, validate, and deploy network configurations across any implementation of SONiC. This fabric manager can configure Clos network topologies for top of rack. leaf/spine, and superspine layers in a modern, flat network like the hyperscalers and cloud builders use.
The Assurance module keeps tabs on key network metrics to enforce service level agreements for network traffic and therefore for applications. The Telemetry Agent module is used to bring in telemetry from switches running SONiC or other NOSes so the complete datacenter and edge network can be analyzed in real time.
And finally, Aviz Networks is offering Level 3 24×7 technical support – including crafting bug fixes on the fly – for all SONiC distributions, backed by teams in four different time zones to provide global coverage across space and therefore time. And it is all priced at under $1,000 per year per switch, with the expected volume discounts of course that drives the cost down by 30 percent to 40 percent across hundreds of switches.
The SONiC support contract is add-on, and there is an option for 8×5 support as well as 24×7 support and the cost depends on a bunch of factors. To give a sense of context, with a Cisco switch, about 60 percent of the cost is the hardware and around 40 percent of the cost goes into the software and support. With SONiC, the portion of that 40 percent that is allocated to NOS licensing goes to zero, and then the support is also cheaper than what Cisco offers. The upshot, says Shukla, is that Aviz Networks can cut that 40 percent of the switch cost relating to software and support in half. Which is why the hyperscalers and cloud builders have been writing their own NOSes and supporting them for a decade and a half.