SDN can turn the network into a big data “curator,” claims Juniper


Credit: Albert Chau

Session Name: Hyperconnected Big Data

Speakers: S1 Announcer S2 Jennifer Lin


Up next we have Hyperconnected Big Data, how SDN will shape sharing ecosystems, presented by Jennifer Lin. She’s the product management lead with Juniper Networks. Please welcome Jennifer to the stage.


Good morning and thank you for having me here to help celebrate the vernal equinox so we can spring into spring together with a thousand of our best Big Data friends. My name is Jennifer Lin and I lead the product management effort. I joined Juniper Networks just a few months ago through the acquisition of a start up in Silicon Valley called Contrail Systems. Juniper acquired Contrail really to look more closely at this opportunity around software defined networking and there are a lot of changes going on there, really driven by a lot of the changed being driven at the application level. Applications are much more mobile, more distributed, more dynamic, and much more virtualized. So, I’m going to spend just a few minutes today talking about Hyperconnected Data, what it takes to enable it and why we think you should care.


Application consumers are really impatient these days. They expect instant gratification anytime, anywhere and this really drives the role of the network to a much greater degree, things are happening much more quickly. The pace at which things are happening is astounding. The number of devices on the network is really exploding. So, we take a look in just one Internet minute and we get a little help from the folks at Intel to look at what happens in just 60 seconds on the Internet. This just provides some context into the challenges at hand and some of the perspectives around it. So, users are really changing on a second by second basis; the types of applications they want to use, the types of profiles they want to expose, the types of answers that they’re looking for. So, you can see here, 13,000 applications downloaded every 60 seconds. 700,000 new Facebook updates, 2 million plus new Google queries. Users are really impatient and so, as we mentioned, we believe that the role of the network is becoming much more critical as the pace of this accelerates.


This is just a snapshot and a diagram of what’s going on the Internet from a network perspective. Obviously, we quickly blew through 4.3 billion unique IP addresses and are onto the next wave, but the registries have exhausted all the IP addresses that originally existed, when in fact it was designed to essentially be unlimited. So, not only are the number of end points growing and we’re really now starting to apply IP addresses to virtualized assets, sub-systems in the data sets, personas. So, the number is really growing, but end points are not just about one-way transactions. Really what the network needs to do is enable much more dynamic and richer interactions and as this data across the Internet is opened up, really the bar is much higher in terms of the types of services that the network needs to expose. So, once again, in just 60 seconds, 639,000 gigabytes of data transferred over IP networks.


The analogy to a very large distributed control system like the human nervous system is interesting. Once again, in just 60 seconds, you have over 12,000 neurons firing, 120 million new cells created, and over 640,000 sensors employed. This type of dynamic control system, the human body obviously does quite well, so we aspire to essentially have the same level of heterogenious transport systems, very distributed transport sensors, but essentially, a centralized control function that essentially keeps everything in balance. One thing though is that, the difference between the brain and the mind can often be debated. Some people use them interchangeably, but the brain is really more about the science of neuro receptors and neuro transmitters and receptors and essentially more of the electrical impulses, whereas the mind is more about intuition and emotion, and learning, and consciousness. Those are the things that if you look at the design from a technology architecture perspective of something like a dynamic control system, yes, of course, you need a controller to essentially feed in system behavior. Yes, of course, you need a feedback loop that essentially ties this all together. Often the challenge of deploying these large scale systems is the intuition part of these systems. How do you inject into this the human behavior and the intuition so that essentially the adoption is accelerate of these large scale systems.


At Juniper Networks, we’ve thought hard about the role at the system level of the networks in bridging what is today, a very wide scale physical network that interconnects all of these IP end points. We want to make sure that as we move forward to enable big data sets and new application developers, that we leverage what we’ve done very well over the last 2 decades to drive a very global hyperconnected IP network. On top of that, we’ve defined how we can essentially create a virtual service overlay where we can dynamically pull in varying network services and increasingly, the network has a lot of context down to a very granular end point level about who is that user? What are they doing? Where are they? Now, we’ve time stamped a lot of this transactional and interaction data. So, essentially a very rich information. As we aplly those services then, the controller plays a number of functions. Number one, we need to federate the control function and avoid any single point of failure, just like the Internet and make sure we have a self-healing system there. On top of that, we need to essentially enable better operational efficiency and get rid of a lot of the manual error prone processes that are present today in human middle ware and highly manual confirgurations.


The key thing and the key role of the network is really a platform for much more dynamic applications. Things like Big Data just for the infrastructure and also at a higher level application set, in terms of driving user behavior. We’re seeing a huge opportunity here to re-position the role of the network as a curator of big data and make sure that that role is easily exposed through much simpler abstractions of the network so the essentially, the expectation is that application developers can just consume services from the network and not have to know about the ins and outs of various protocols.


The role of the network is interesting because the network is the only thing that’s globally pervasive and incresingly with mobile data sets and billions of sensors out there. It is the thing that essentially uniquely knows a lot of the contextual information that is required to drive insight back into the system. So, notions around domain insight and context coupled with a lot of instrumentation down to individual end points about time stamps, about location, about identity, about presence; this is all rich context that essentially helps drive system level behavior. It will also give rise to new types of business models like collaborative exchanges, where producers and consumers of information can interact across heterogenious data sets without worrying about the technoloy architectures in between.


So, in the brief period that I had today, a message that I really wanted to drive here was really how do we allow application developers to empower their data at scale and really drive and assume that over an open flexible standard space platform, we can drive much richer insights. The role of the network as you can imagine is changing quite heavily and the pace of innovation for Hyperconnected Data is really astonishing. We look forward to playing a large role here and thank you for your time today and Happy Spring.


Jeroen van Bemmel

Mike is not the only one who is confused. Why do you switch to defensive mode, quoting Juniper’s strategy (in very vague terms, adding to the confusion), while all he’s saying is that there is a need for more common language and concepts? Do you disagree?

Jeroen van Bemmel – Alcatel-Lucent

Michael Bushong (@mbushong)

[full disclosure: I work on the vendor side in the SDN space, and I was a long-time Juniper employee driving SDN at Juniper before it was called SDN]

I absolutely agree that there is a need to focus on abstractions. It is great that we are working our way out of the minute details of the underlying protocols. I wonder if the abstractions mentioned here are still a bit off though.

It makes perfect sense for networking companies to view the network as the center of the universe. And in many ways, because it provides connectivity across all of the infrastructure, it’s actually true. But when data resides across parts of a network (or worse yet, across multiple networks), the abstractions need to be more than just decoupling hardware and software. There needs to be a common data model that multiple vendors (not just networking) can use. I don’t see that work really happening here.

And I will give a shout out here to Tom Nolle, who would likely point out that Juniper continues to talk NFV while referring to SDN. The two are related, but they are different. We really need to get more clear in how we talk about things because it’s confusing as hell to customers, which hurts all of us on (on both the vendor and customer sides).

Ankur Singla

Definitely you are quite confused about Juniper strategy, surprising given that you were part of the company till recently. The strategy encompasses the entire spectrum of networking space – Enterprise Private Cloud, SP Access & Aggregation layer, SP Core, Public Cloud/SP VPC.

–Ankur Singla, Founder/CEO – Contrail Systems

Comments are closed.