The appliance model, where the hardware and software were tightly controlled by a single vendor, held sway in the datacenter for decades. But that top-to-bottom stack been peeling apart, oddly enough because it was never established on the PC, but particularly as Linux emerged as an open, cross platform operating system that was eager embraced on servers, first by the HPC community and then by the hyperscalers and the cloud builders.
Even with Unix servers, which were open in terms of supporting common APIs and providing a modicum of cross-platform compatibility, the operating system and related software stack was generally made by the same company that designed the processors and the hardware system that used them. Linux and Windows Server, which jumped form the desktop to the datacenter after decades of hard pushing and platform buildout by Microsoft, provided portability and superior price/performance for many (but not all) workloads, and this finally broke the server software free of the server hardware.
It is not a foregone conclusion that this can happen in the datacenter switch and routing markets. Many have tried with open source – and generally Linux-based – efforts, and not all have failed, none can be called a resounding success, either. The reason, one could argue, is that open networking was never really about open source, per se, as much as it was about creating good, portable software that had a dedicated team of people working behind it. That’s really why Windows Server and Linux were successful on servers – although they had very different development and commercialization models, of course.
This is why we say that Arrcus, an upstart switch and routing operating system provider, wants to be the Windows Server, not the Linux, of network operating systems. Arrcus has no interest in opening up its software whatsoever, and neither do its venture partners. But with ArcOS, the company absolutely wants to create what it hopes will be a universal substrate that can do both types of networking jobs on all manner of merchant silicon, providing that portability and compatibility that Linux and Windows Server have brought to compute.
Devesh Garg, co-founder and chief executive officer at Arrcus, and Murali Gandluru, vice president of product management at the company, talked about the potential for a cross-platform network OS at our recent The Next I/O Platform event in San Jose. The obvious question is why does ArcOS have a shot at breaking the hegemony of the appliance model in networking?
“In terms of why now, I think it is the aggregate market and the pressures that are forcing people to rethink how they are doing things,” Garg explains. “We live in a hyperconnected world, depending on who data you believe, there are going to be 50 billion to 70 billion interconnected devices between 2025 and 2030. The Internet is getting flattened, and then there is the advent of 5G, which is going to bring 10X the number of subscribers per square kilometer compared to 4G. So I don’t think I need to make the case too strongly that the world is hyperconnected. But above and beyond all of this connectivity, there are bandwidth and latency issues, and all of this needs to be done ultimately at a lower price point.”
This is just one set of vectors driving the industry towards smashing the appliance model in switching and routing and adopted a more disaggregated networking model. The other vectors that are part of this overall historical trend, adds Garg, are the tremendous amount of innovation that is being done my the merchant silicon vendors and their switching and routing partners, and this includes, as we have talked about before, the addition of a substantial amount of routing capability into switches using those merchant ASICs. But perhaps more importantly, as far as Garg and Gandluru are both concerned, until the advent of ArcOS, there was not an open alternative that was as good as IOS or NX-OS from Cisco Systems, JunOS from Juniper, or EOS from Arista Networks. Being open is not enough – a network operating system has to be portable across different ASIC suppliers and it has to be jammed with features and deliver on performance.
Just like Linux did not do when it first came into the scene in the middle 1990s, but it certainly eventually did do in servers thanks to a lot of help in the open source community – spearheaded in many cases by the HPC and hyperscaler communities that benefitted first and most from the work to make Linux scale further and run faster.
One of the edges that Arrcus has – that many of these open source NOS projects did not have – is a deep and broad experience in routing, which in many ways is a much harder piece of work than creating a switch operating system.
“Networks have moved away from L2-centric environments to routing-centric and highly programmable models,” says Gandluru. “As that happened, it became imperative that a common, routing-centric design applies from the datacenter all the way out to the edge. Fundamentally, a routing centric, standards based, highly programmable approach enables a scale up and a scale out design, depending on the part of the network. When you combine that with a composable, microservices-based operating system, you now have the ability to target different parts of the network with focused, composable software – all routing centric.”
If ArcOS takes off, as it has a very good chance of doing, it will be interesting to see what the big switch and routing incumbents do with their software. They do not want to sacrifice the huge profits and account control they get from this software, and they may end up no better than the Unix operating systems that have been relegated to history.