Sponsored By

Pradeep Sindhu, Juniper Networks

"Sometimes it appears to be like a slow bicycle race: Who can go the slowest without falling down?"

April 26, 2004

11 Min Read
Pradeep Sindhu, Juniper Networks

{Image 1}Pradeep Sindhu, the founder, vice chairman, and CTO of Juniper Networks Inc. (Nasdaq: JNPR), has come a long way from his days at the Computer Science Lab at Xerox PARC, the birthplace of many a networking technology, including Ethernet and the first optical-cable-based LAN.

In 1996, Sindhu left Xerox after he cooked up the idea for a high-end router company. Sindhu was joined by Scott Kriens, who became Juniper's CEO, and the rest is history. Juniper was up and running in a couple of years and staged one of the most successful technology IPOs in history in 1999.

We met with Sindhu at Juniper's headquarters in Sunnyvale, Calif., in early April. A couple of weeks later we enjoyed hearing Sindhu speak at the Light Reading Live investment conference in New York, "The Telecom Recovery: Opportunities Amid the Chaos."

We've combined and edited the contents of these separate discussions and presented them here in this Q&A.

Sindhu's main role right now appears to be crusading for Juniper's "Infranet" initiative. Yes, the name sounds like something from the Jetsons. Some have even called it a "marketecture." But Juniper's dead serious about this Infranet thing, so we thought it was important to find out what it's all about.

The idea is to get any device to connect to any other with the appropriate speed, QOS, and security. The necessary tradeoffs among four dimensions – security, speed, reliability, quality of service – need to be handled according to the needs of each particular application. (Connectivity is a fifth dimension but isn't as problematic as the other four.) The challenge is to build a network that can simultaneously weigh all those factors and build connections for applications that suit the range of needs across this four-dimensional spectrum.

So, what Juniper says the telecom industry now needs is a bulletproof packet network that can be adjusted to applications on the fly, depending on where they fit in the four dimensions.

It's a tall order. The stuff we have today won't cut it – as Sindhu freely admits. Sindhu and Juniper are suggesting the industry should put some serious planning into this, with a focus on some new standards (see Juniper Does Vision Thing).

Will it work? Who knows. In order to make it work, Juniper needs to move many of the largest service proviers in the world to support its efforts. Find out more about what this is all about in the following pages:

  • No Nap for Apps

  • Surfing for Surgery

  • Standard Matters

  • Bellheads vs Netheads

— R. Scott Raynovich, US Editor, and Craig Matsumoto, Senior Editor, Light Reading

Light Reading: At our upcoming conference we intend to talk about the most exciting new applications. Do you have any thoughts?

Pradeep Sindhu: You never really know what the best applications will be. There's a very strong analogy between the networking world and the computing world. When computers were first invented, people had a very one-dimensional view of their capabilities. But it turned out that computers were universal, in the sense that you could program them to do almost anything. Every time you upped the speed by a factor of 10, the applications that fell under the umbrella of computing were completely unanticipated. More and more applications started falling under that umbrella.

I think that networking – and particularly IP-based networking – has exactly the same flavor as computing. They share one essential thing: The higher the performance of this any-to-any network that you provide, the more applications that will actually fall under its umbrella. The network and the computer are like yin and yang. They're not separable. We tried to separate them in industries, but fundamentally they're not separable. They're co-dependent.

Within a very short number of years, any entity that connects to the network will have a computer in it. This is sure... cell phones, even the smallest devices. Everything will have a computer. So when you talk about the network connecting people – yes, that's where it is ultimately, but our representatives actually sitting on the network are computers.

LR: So you don't see any particular killer applications? PS:If I rolled back time 45 years, and people ask about killer applications for computers then, I'd say the killer application was computing ballistic tables for lobbing shots to enemy targets. Today, I think that most of the applications for communications haven't yet been invented. It's because the number of things that it is possible to do grows very rapidly with the underlying infrastructure

For computers, for example, in the 80s the killer application was desktop publishing and document handling. Today it's way more wide than that. And the interesting thing is, the power of computers is multiplied by some very large magnitude when they're connected, compared to when they're not. And so this is what leads me to the intuition that the same phenomenon is happening here, except on a much, much larger scale.

So, I don't have any favorite program.

[Editor's note: To illustrate his concept of four axes – bandwidth, security, reliability, and quality – Sindhu points to the example of "virtual surgery." To our knowledge, nobody's doing virtual heart surgery yet, but at least Pradeep's thinking about it. One hopes it doesn''t turn out to be a killer app.]

LR: So, when are we going to have virtual surgery?

PS: That's just an example. You, know, it may be fundamentally impossible to simultaneously ask for very high connectivity and very high security and reliability. Something tells me that that might be true.

So, this is why the notion of virtual private networks has come about, because I can certainly simulate a private network on a public network. For that private network, I'm able to provide security and availability because I can reserve some resources in the network. But if I'm not allowed to reserve resources, and I can manage to talk from anywhere to anywhere, I actually don't know how you would guarantee anything. In order to guarantee something, I think you have to restrict the set of people temporarily, just for that one area when you're talking.

LR: A lot of investment has been made in enterprise networks. Do you think the focus going forward is going to be on consumer networks, on being connected to the home with applications like video on demand?

PS: I wouldn't say that. I would say that different customers are focused on consumer applications and other ones are focused on enterprise LANs. A very broad area of customers.

LR: But it's been very hard for service providers to make money from new data services, especially when many businesses already have data services.

PS: Let's ask the question: What is it I really want, with respect to information? What I'd like is to have access to all the information I want, independent of where I am, at exactly the same quality I would get at my desktop – and without having to know and learn details of the technology I'm using.

If you evaluate things on that method, we're very far from that. Try using your cell phone to browse, right? It's like pulling teeth right now: partly because the screen is small, partly because there are a lot of gaps in the technology. We're very far away from being able to access information anywhere while being guaranteed that it's secure and reliable and can run different applications. We're very, very far from there. It's partly because of the hodgepodge of technologies that have been put together.

LR: You've mentioned two standards that you've proposed as part of your Infranet intiative. Could you explain these?

PS: There are two particular interfaces that absolutely need to be standards and for which there need to be open standards. These are the Inter Carrier Interface [ICI], which is the interface between one carrier and another, and the Client Network Interface [CNI], which is the interface between a client and a network.

The CNI is something like the UNI that ATM had, but it is a lot more comprehensive. The reason it needs to be more comprehensive is, imagine if you have this application span, which covers a broad range of applications, there has to be a way for applications to tell the network what it is they want – across the spectrums of quality, reliability, and security. Similarly, with the ICI you can communicate across the network between two carriers. You need to carry the same information you have in the CNI, which says exactly what the application needs. There are no standards here. The ICI also has to have a way for carriers to settle charges with each other. This is a very important piece that is missing today.

The two standards, CNI and ICI, are very important to bringing about this world of the Infranet. The existing standards are not sufficient. Juniper has issued a call to action to help us develop these open standards so that we can actuate this vision of the Internet.

LR: You're trying to influence an entire standards movement. Are you encountering resistance because this may favor the incumbents?

PS: It doesn’t favor anybody, because open standards are interesting because everybody can participate. Clearly, incumbents have a leg up, but incumbents who base their equipment on prior standards do not have a leg up. Actually, open standards benefit the insurgents.

LR: But if the service providers all use the same standards, won't it become harder for them to differentiate among themselves?

PS: No, because if you want to build the network that has all these attributes, it is a very difficult problem. Not everybody’s going to do it as well.

LR: How does Cisco’s cooperation factor into this?

PS: If Cisco were to embrace this, they would have to build equipment that has the appropriate mechanisms that can support a broader range of applications. Let me give you some examples: mechanisms for queuing, mechanisms differentiating between packets, mechanisms for deep packet lookup without compromising the rate at which the machines work.

Today, the machines that Cisco builds do not have these capabilities. They have it in software, but if you try to exercise the capabilities, the performance of the machine drops. Well, that’s not a realistic way of having them. It’s like saying you can have this or this but not both.

LR: Have you had any response from Cisco?

PS: All I will say to that is: Stay tuned.

LR: Many of these standards exist in the circuit world. Won't this turn into another episode of the Bellheads versus the Netheads?

PS: When I talk to our customers, there has been virtually universal agreement that something like this needs to happen. There's also very strong agreement this needs to be standards-based. What you call Bellheads...

LR: You don't call them Bellheads?

PS [pausing, but ignoring]: [We're talking about] a world where voice was done based on worldwide standards, and, in fact, if you did not have those standards, I would submit that this voice network we have today would be nowhere near what it is. Standards that defined not only the technology but also defined ways that carriers could make money by carrying traffic.

LR: I'm just playing devil's advocate.

PS: We've not gotten very much pushback at all.

LR: We were just meeting with Tony Li [former Juniper employee and Procket Networks Inc. founder, whose Interview will be posted soon]. We were talking about how in the large carriers you have circuit guys or you can talk to the packet guys. They're buying completely different products. We were asking him if that still exists, and he said it's as divisive as ever.

PS: I think this debate between circuit versus packet is effectively over. You still have a few religious zealots, but I think that fundamentally, that debate is over. The reason there was a big debate was, first of all: Could you actually support a broad array of applications on a packet infrastructure? At one end of the debate, people would say it's not possible because of delay, jitter, and such. It's unguaranteed.

The other side of the debate was: We have circuit-type things where you have very, very bursty traffic, and no one knows how to build an economic network given those conditions. If you follow my remark about every device at the end of the day being a computer, even if it's a human being talking, all traffic is going to have that flavor now. Given that, how would you want to design the network in an economical way? It's got to be a packet-based infrastructure.

So, for most of the people that I've talked to, there's no debate any more. It ended five years ago.

LR: A lot of the problem is: How fast can they move?

PS: Yes, sometimes it appears to be like a slow bicycle race: Who can go the slowest without falling down?

Now, what I want to be clear about is: Don't underestimate the difficulty of doing this [the Infranet]. This is a phenomenally hard thing. But I also want to say, at the same time, it's a very worthwhile thing.

Read more about:

Subscribe and receive the latest news from the industry.
Join 62,000+ members. Yes it's completely free.

You May Also Like