Juniper Anywhere: Where Will the Truth Live?
September 2007
How does one of the industry's most important leaders view advancements in connectivity? Zeus Kerravala and I sat down recently with Pradeep Sindhu, vice chairman and chief technical officer of Juniper Networks. Pradeep and his experts are thinking hard about the impact of ubiquitous connectivity on enterprise data centers—their computing, their staffing and their role.
Emily Green: Yankee Group's perspective on the world—what drives our research agenda—is that we are moving rapidly to a state of ubiquitous connectivity, where all things and people will be connected with each other.
Pradeep Sindhu: And it's inevitable. The reason for this is the network effect. The fundamental value of the network comes from the number of things connected.
Emily: Right—it was true when the number of computers on the net increased, then it became true for the number of people on the network. Now it's about things on what Yankee Group calls the Anywhere Network™.
Pradeep: We are indeed driving to a universal network. And with that, the quality of the experience, with regard to the access of information wherever it is, should look, feel and smell the same anywhere you are.
Zeus Kerravala: But the ways users access that information have to be appropriate to the device and the network.
Pradeep: Exactly. But there are really two forces at work in this move to ubiquitous connectivity. One is the network effect—encouraging any-to-any access. The second force is computing. All useful computing activities will be connected to the network.
Communications and computing are like yin and yang. A great parallel from the world of biology is the evolution of the human brain and the evolution of language. When you look at brain size, and the desire and ability to communicate, you find that each one affects the other one in a positive way; they're mutually reinforcing.
If that's true in the relationship between computing and connectivity, then the evolution of a first-class network will have a profound effect on computing.
Computing will move to two main models: first, computing that is close to the user. This computing activity will be optimized to the use and the user. It will demand new things from the devices at the edge, such as dramatically lower power use and longer battery life, so we can put together big sensor networks, for instance.
In the heart of the network, there is an opposite effect on computing. Rather than special-purpose computing designed explicitly for a particular use, we’ll see the continued optimization of computing for scale using general-purpose compute power.
Zeus: Today's corporate computing environments have become pretty distributed. Problem is, once that happens, it becomes very hard to manage, less effective, more expensive. You have to start thinking about how to orchestrate it all.
Pradeep: The data centers of the future will be much larger, and thus economically much better than today’s. So computing will probably not be done best by distributing it into teeny little pieces.
That's because how tightly coupled the computing elements are affects the programming model you have to use. In the most extreme distributed approach, producing designs like Connection Machines, the impact is to make programming really hard. Most programmer brains don't work well enough to do that successfully.
Right now I think there is an impedance mismatch between computing and connectivity. There'’s an order of magnitude efficiency improvement available. Whatever we design at Juniper for this problem needs to take this into account. The industry is currently partitioning the complexity of the issue all wrong, up and down the stack—
Zeus: Meaning, what goes into hardware and what goes into software?
Pradeep: Right. The industry doesn’t have good taste on this matter.
I think there are going to be two kinds of data centers. One designed for highly specialized applications—for instance, oil exploration or information search.
The other is the more general data center. Small and medium-sized enterprises (SMEs) will rent data center space from AT&T and others. There the computational infrastructure will evolve every 6 months, and the provider can afford to keep up with it because of the scale and client base, and SMEs can finally benefit from a current IT environment.
A big driver of scale in the enterprise environment today is software-as-a-service. End-user devices will ultimately not have what we think of today as persistent storage.
Emily: If the network is always available, and the application is in the cloud, why shouldn't most of the data be there, too?
Pradeep: End-user storage will be temporary storage; more like a cache. I'd define it as "if I lose it, I don't have to care." The truth—well, the truth will live somewhere else.