SDN Networking with Cumulus Linux
Today, Cumulus Network emerged from stealth and unveiled Cumulus Linux with a couple of impressive, early adaptor customers including Dreamhost and Fastly and investors Battery Ventures and Andreessen Horowitz. They also alluded to Amazon as a customer via GigaOM and blog post by James Hamilton a well respected Distinguished Engineer at Amazon.
Cumulus’ value proposition is that customers need a Linux of networking for to a) reduce costs; and b) provide greater agility. It’s important to note that today’s announcement is focused on a networking OS for bare metal switching hardware and not a vSwitch. I’d be surprised if we don’t see a software only version announced for servers shortly.
I haven’t been personally briefed on the announcement – though given my experience as CEO and co-founder of Pareto Networks – I have scar tissue from this approach.
At Pareto we had full Linux OS (including many of the same features as Cumulus) and our software ran on bare-metal hardware from ODMs (Accton was our initial partner) as well as bare metal switches and wireless access points from two of the top three networking companies. Our initial solution targeted at remote office/branch office and we also had shipped a software-only edition that ran on all of the leading hypervisors and could be automatically provisioned as needed. This was in 2008 and Pareto was acquired by Aerohive in 2011.
Bare Metal Networking: Experience
At Pareto we learned a number of things about selling an OS on bare metal:
- Most customers care more about ‘one-throat-to-choke’ for the entire solution than they do about vendor lock-in or slightly higher hardware prices. The result was we had to pick one ODM for each model and sell a bundled solution that includes hardware, software, and support – to move beyond trial deployments in the enterprise.
- If the incumbents want to win a deal from you on price – they can and will. The paradox of bare metal is that it’s produced on the exact same production line as many products from the incumbents. These incumbents also happen to be the largest customers to the chip manufactures and the ODMs. The result: due to economies of scale, the incumbents can always beat bare metal on price, IF, they choose to. Hence, why I’m skeptical anytime a startup comes to tell me that are going to be cheaper (or more ‘afforable) than XYZ networking vendor. They attempt to make the argument about software and support is cheaper – the reality is in most case bare metal simply represents an expense shift across two vectors a) between hardware, software, and services; and b) over time. More specifically – I have yet to see a business case where the cost, cheaper, more affordable argument holds over a 5 year period (though if you have one, we have client’s who’d love to see it).
- You must have sufficient differentiation in software to add new value. This value proposition has legs – if you can provide greater flexibility and agility (which means you can do more with less) via software then you have something of customer value. At Pareto we did this by marrying the network functions (and OS) to a cloud service (ala Meraki). This is reason why we had a major financial and 3 leading hedge funds as our first customers – we solved a unique problem with our software. Assuming Cumulus delivers on this — then they have something that’s meaningful.
Cumulus and Cumulus Linux OS Matter?
Does the Cumulus launch matter, to whom, and what can we take away – our perspective and questions:
- Cumulus does matter – though likely to a niche (albeit an import niche) to such as cloud data center operators who are likely to build their own switches. Cumulus’ hope is that traction with these customers (who are similar to the Nicira customer list) will drive someone like VMware to acquire them as their network operating system. This team is also rumored to have worked at Google on the Pluto switch – so Cumulus’s technology should be a credible for this market segment. I have a hard time seeing Cumulus going beyond the lab in enterprise in the short term – see point #4 below.
- Hardware still matters – Our market research shows that we expect significant network datacenter spend to be attached to hardware. Programable Linux based OS that can run on commodity hardware supports our assertions.
- Cumulus is not alone with Linux OS for networking. Arista’s EOS, Pluribus, and Pica8 all make similar claims and varying business models.
- Who’s responsible for support? If I’m Citibank and I’ve bought Cumulus + Quanta – who do I call when I have a problem? How do I know you have the support processes and capabilities in place? If I call Cumulus and we find out there’s a bad port on the switch, how do we deal with that?
- Support DNA is needed – looking at the DNA of the company – it’s a product/software experienced team looking to monetize via services model. In my experience – they need to quickly ramp services expertise to be appeal beyond a handful of highly technical customers.
- Can anyone implement Cumulus in their network? Short answer is not today – if we look at the ODM hardware listed on James Hamilton’s blog – only the Quanta QCT T1048-LB9 (48x1G & 4x10G), and Quanta QCT T-3048-LY2 (48x10G & 4x40G) hardware even have a web page or data sheet available for download in English. Meaning, today, you need a personal relationship with the other ODMs just to learn about these platforms.
- Technology Partnerships: This is the one area where the announcement lost credibility with me. The partner list is longer than Richard Petty’s NASCAR sponsor list – and while it’s good to show that you have friends and big brothers supporting you –most large companies are not equipped to have meaningful partnerships with that many technology partners – let alone a startup; which makes it challenging to understand if there is any meaningful integrations or partnerships.
- Open Source: Cumulus claims to be open sourcing their OS, however I was unable to find a download from their site. As you know — we are SDxCentral are skeptical of single vendor Open Source solutions — and especially how it doesn’t not protect you from the same vendor lock-in these vendors are attempting to rally against.
Like we’ve said before – this is an interesting time to be in networking and look forward to seeing where Cumulus Linux takes off. We are encouraged with the concept of bare metal switches and see a segment of the network that Cumulus can address now the question is – can they execute beyond cool technology while avoiding the traps of perceived open source software.