The world of SD-WAN has been a fast paced one in the last couple of years, and there are many flavors of the technology out there coexisting under the same umbrella. Yet each has its own unique approach and set of capabilities. With us today to delve into the subject is Andy Gottlieb, VP of SD-WAN Solutions at Oracle Communications. Andy was a co-Founder of SD-WAN pioneer Talari Networks, which Oracle bought late last year and integrated into its Communications business.
TR: What is your current role at Oracle and what was the journey you took to get there?
AG: I was one of the two co-founders of Talari Networks, the first SD-WAN company. We did SD-WAN before it was called SD-WAN, before even software-defined networking came out. We were a small start-up and we had the best technology, but we just didn't have the global reach. Becoming part of Oracle's communications unit a few months ago gave us the global footprint we were looking for as well as a significant expansion of our R&D capabilities.
TR: What type of SD-WAN does Oracle offer? What type of solutions are you targeting?
AG: The rise of the Internet revolutionized IT except for, ironically, the enterprise wide area network. WANs have been the domain of expensive, private "overengineered" networks, and the price per bit of MPLS is at this point 100 times the price per bit of internet. We realized that by developing technology to make the internet more reliable than MPLS, we could get much more bandwidth at a fraction of the cost. At one point, we described this as Redundant Array of Inexpensive Disks (RAID)for WANs. With RAID these guys at Berkeley in the 1980s looked at those big expensive mainframe single-disk storage systems and had the clever idea to wrap a layer of hardware and intelligent software around multiple small hard disks from ‘toy’ personal computers. They created a storage system that was both much higher capacity, much lower cost and more reliable, and quickly revolutionized the storage business. It’s similar for WANs, and the way we solved it is to continuously measure all of the different paths between any pair of locations, whether enterprise site-to-site or site-to-cloud. We can take advantage of MPLS but also any public internet connections, cable, 3G, 4G, LTE, Metro Ethernet, whatever is available. The joke is that we can use two tin cans and a string as long as it supports IP. By measuring all that traffic, we know what the loss, latency, jitter and bandwidth utilization is of each of the different paths. We can then make packet by packet forwarding decisions and react literally in fractions of a second to problems happening on the network. That's the key to making this multi-link fabric more reliable than the best MPLS even when using very inexpensive consumer broadband connection.
TR: What is it that makes MPLS so much more expensive?
AG: 25 years ago when Frame Relay came out, it was actually groundbreaking price/performance. It replaced all these point-to-point leased lines and was really the original data cloud network. Everyone had a single connection into this cloud, and that way you could connect to everybody else. So within a very short time, the uptake was faster even than the internet itself. When MPLS came out as the successor to Frame Relay, it was better supporting real-time traffic like VoIP, and it was simpler to manage. But pricing remained high. Because unlike everyplace else in technology where when you cut the price in half consumption goes up by a factor of 10, with Frame Relay and MPLS-based private WANs that never happened. The carriers came to realize that cutting prices didn't add volume. So while Internet bandwidth followed Moore's Law down the cost curve, premium MPLS prices only went down 10 to 15 percent a year. And so what 15 years ago what may have been a 20x or 30x gap is now a 100x gap.
TR: So what do you do differently than other SD-WAN technologies?
AG: Most of the competition doing SD-WAN basically view it as two things: a way to consolidate multiple functions in the branch and way to do simpler configuration and management of the WAN. We see those two things as only maybe 25% of the problem. If you don't fix the reliability and quality of experience problem, then it's going to be difficult for customers to migrate away from MPLS, or even to augment MPLS with the internet connections. And it's going to be particularly difficult to do reliable, high-quality of experience applications in the cloud. Other SD-WAN technologies will say, as an analogy, that there are three ways to get from San Francisco to San Jose, and highway 101 is usually the fastest and highway 280 has more lanes. So they put the most important traffic on highway 101 and put the bulk traffic on highway 280. And when it gets bad on 280, they can just move everything to 101 and vice versa. With our technology, when there's a problem on the road we detect it very quickly. We route stuff around in less than a second. For most, one session goes this way and one session goes that way, and that's all you can do. We can make the decision packet by packet, fix problems automatically – enabling the network to “self-heal”, and then we just send you an alert about what the issue is, rather than make you troubleshoot the WAN yourself.
TR: To what extent do you leverage AI and machine learning to make those decisions today, and what will you be able to in the future?
AG: In my computer science days from a long time ago, we’d say that AI is AI until everyone understands it and then you don’t call it AI anymore. What we have is a site-to-site machine-learning system that is doing a very specialized form of big data. We analyze all the information about packets traffic across multiple paths between locations, and based on that we make decisions in real-time where traffic should go. That part of it we've been shipping almost since our beginning as a startup and of course as part of Oracle for months now. We can also utilize additional levels of that machine learning by aggregating all this information at an even higher level across multiple sites and taking advantage of cloud computing for things like capacity planning. But fundamentally it is all based on our continuous, unidirectional measurement technology. We actually don't have an exact one-way measurement; we merely have a really good estimate. But what we do have is a perfect estimate of the relative one-way times of each of the different paths and that's what you need to make forwarding decisions.
TR: What kind of enterprise customers do you see the most demand for this SD-WAN technology from?
AG: We have been very successful, for example, with emergency 911 systems. In the US, 911 systems are run by states and county governments, and unlike the federal government they don't have tons of money. The first one, and now a longstanding customer, is Maricopa County, Arizona. They actually used our SD-WAN to get rid of a Frame Relay network and replaced it with consumer broadband. They were looking to save money and improve quality at the same time. Our SD-WAN is used by ~70% of the state of Texas for E911. When the big hurricane hit Houston in 2017, we called them the next day to make sure everything was okay. While all the wired networks connections were underwater, their wireless ones were still working and they hadn’t dropped any calls. We’ve also done well with banks and credit unions, who care about reliability and security. All SD-WANs do low-level data security just fine, but they also wanted the high quality of experience and application performance predictability – and cost often becomes an issue as well. As you want to roll out next-gen applications like video to ATMs, for example, then you need a lot more bandwidth. But most people don't want to pay eight times the price to get 10 times the MPLS bandwidth. So they typically augment their MPLS connections with internet connections and use our technology to get better reliability than they had before along with more bandwidth and lower costs. We also have a lot of offshore oil rigs, cruise lines, government and military applications. The short answer is we're a natural fit for anyone who's got real-time or mission-critical applications.
TR: Do you prefer direct sales to enterprises, or do you use telcos or channel partners?
AG: We have a “high touch” sales team but, but most of our business goes through classic reseller and master agent/subagent channels. Channel partners are trusted advisors to their customers, and some of them really do their homework and look at technologies. We think we’ve got the best technology out there so if they carry five different SD-WANs we don’t mind because we win a disproportionate share of the time.
In North America, we don't use telcos as a distribution channel at all, although we do have some telcos overseas that are Oracle partners that we're working with.
TR: What do you see is the biggest challenge ahead for enterprise WANs or the SD-WAN? What should be keeping people up at night?
AG: The big challenge for enterprise WAN administrators and network managers is how to safely migrate the applications to the Cloud. We would argue what you need is fail-safe technology to be able to do as well site-to-cloud as MPLS is giving you site-to-site. Most other SD-WANs make network management easier but can’t deliver that reliability and QoE. And so I think that's the most important thing because otherwise you are doing the thing that customers hate: taking two steps forward, but one step back. It’s the quality of experience challenge that people have to worry about.
TR: Thank you for talking with Telecom Ramblings!
If you haven't already, please take our Reader Survey! Just 3 questions to help us better understand who is reading Telecom Ramblings so we can serve you better!Categories: Industry Spotlight · SDN