This month's winner of the Network Innovation Award is Arista's EOS+ network operating system, which makes the programmable switch something that's finally accessible to mere mortals (and enterprise IT pros alike).
EOS+ consists of several elements, chief among them being a formalized software development kit that can be used to develop custom applications for Arista switches. It also comes with three pre-built applications: ZTPServer for automated network provisioning, EOS Telemetry Suite for Splunk Enterprise integration and EOS DirectFlow Assist for security acceleration. These apps can be customized with the help of Arista's professional services arm, EOS Consulting Services, if an IT department lacks the expertise to do so themselves. Apps can also be tested in vEOS, a virtual switch test environment.
Arista founder and CTO Kenneth Duda recently spoke with SearchNetworking about EOS+ and its applicability to the enterprise.
What led to the development of EOS+?
Kenneth Duda: Arista has been focused on switch software -- specifically, the programmability of that software -- from the very beginning. Our vision from the founding of the company was that networks would work better if the software was open and accessible so the customer could modify and extend that software to best meet the needs of their specific operating environment. We had all the hooks in there from the very beginning, and what we realized was that what we had built was good for maybe the three customers in the world who had the sophistication, the development capabilities, the technical wherewithal and the organizational imperative to really harness the nuts and bolts that we had provided to them. But for most customers, this was really just out of reach.
What we tried to do with EOS+ is take the advantages of programmability that, until then, had been harnessed by only the world's largest Web 2.0 companies and make them available to a broader set of customers. We wanted to really enable enterprises -- whose main focus might not be network technology but are still quite sophisticated in the way they approach their infrastructure -- to benefit from the openness and flexibility of the EOS software architecture.
Kenneth Dudafounder and CTO, Arista
I understand you were looking at some of the applications your larger customers had already built with EOS. Tell us more about that.
Duda: The nature of networks has changed as more applications are delivered as services across the wide area network. For example, it used to be that an email program ran on your local machine and would occasionally fetch new messages from a mail server somewhere. Email today is widely consumed as a Web application hosted in a remote data center and that has changed the nature of the traffic on the wide area network. [Changes like this have] dramatically expanded the volume of that traffic and it created challenges for our customers who are trying to figure out how to meet that demand at a reasonable cost point.
One company that is particularly challenged by this is Netflix, which delivers mountains of content over wide area networks and needs to do so in a cost-effective manner. Netflix has taken advantage of our switches and a specific feature called selective route download, which enables Netflix's software to control which specific Internet routes get loaded into our hardware routing tables. And by using this mechanism, they can get the same effect as a wide-area, carrier-class Internet router but using our much more cost-effective data center switch hardware platform.
Another company that's done something similar is Spotify, which delivers large amounts of content over the Internet. They've developed something they call a software-defined Internet router, or SIR, which is based on our hardware platform and also takes advantage of EOS programmability to manage the routing resources in a way that optimizes for Spotify's actual delivery patterns.
These are examples of the kinds of things some of our most forward-thinking customers have done that we've then been able to help enterprise customers take advantage of through EOS+.
Most enterprises don't have Netflix-like demands. How can applications like these benefit them?
Duda: Enterprise traffic patterns are also changing, thanks to cloud. And while they may not be facing it quite on a Netflix scale, they are very much facing the reality that more traffic needs to traverse the wide area network once you move to a cloud-based model for provisioning your applications. Trying to do that efficiently without spending unreasonable amounts of money becomes important for enterprises as well.
As you mentioned, Arista has been focused on building a programmable switch from the start. But most networking vendors are getting into that now. What makes EOS+ different?
Duda: Our competitors have realized this is important, but the problem is they didn't design their platforms to support this. When I first proposed -- in our little startup company 10 years ago -- that we leave the software open so customers could add their own software to the switch, my initial team was skeptical: 'Really? You're going to let customers put their own software on? What if something goes wrong? How will we ever support them? Networks have got to work. This is crazy!'
I said, 'No, guys, I don't think this is crazy. I think not all customers are going to do this. People who have the technical understanding will be the early adopters here, and based on their experience, we can figure out how to structure this so that other companies can take advantage. We're a startup and we can take certain risks, so let's design our architecture from the ground up under the assumption that third parties will be adding software to our boxes.'
We went ahead and did that, and there's a whole bunch of decisions you make along the way [as a result]. For example, if you base your network operating system on Linux, what sorts of changes do you make to the underlying Linux system? If you're simply viewing Linux as an internal component the customer will never see, you're likely to make a lot of changes. But if you're taking the view that the system is supposed to remain open and programmable, then you take a different path. You take a path that's maybe a little more work and a little slower, but … you put a lot more emphasis on maintaining compatibility and interoperability with the Linux world than you would if you designed a closed system.
What our competitors are struggling with is the fact that their systems have 10 or 15 years of engineering behind them where the assumption from day one has been, 'This is a closed system,' so they've made no end of hacks and modifications in their network operating systems. Now, they're facing the reality that they can't just run off-the-shelf open source software on their devices; it has to be ported. The barrier that creates for third parties is something they can gloss over in their marketing slides, but when customers pull back the curtain and try to figure out what's actually required to try to get something to work, the difference between having a true Linux environment versus something that's been, let's just say 'modified extensively' -- that is the polite phrase -- will become very apparent.