Q & A With Juniper Networks CTO Pradeep Sindhu

Juniper's CTO talks about the company's Infranet Initiative.

June 24, 2004

13 Min Read
Network Computing logo

Pradeep Sindhu, Juniper Networks' vice chairman, founder and chief technical officer, is also the public point man for the company's "Infranet Initiative," an ambitious industry-coalition effort to build standards that will bring private-network performance to the IP/MPLS Internet core.

In a meeting last week at Juniper's headquarters in Sunnyvale, Calif., Sindhu provided some more background on the reasons for the Infranet Inititiative, where it stands now, and whether or not Juniper has asked competitor Cisco Systems to participate (they have). In today's segment, Sindhu explains the rationale behind the Infranet Initiative, and what problems it seeks to fix, among other topics.

Advanced IP Pipeline: When you first announced the Infranet Initiative, it seemed to lack a supporting cast. What is the current status?

Pradeep Sindhu: We had kind of a preliminary meeting at the GSM Conference in Cannes [in February], where we privately met with people to share the idea of what we wanted to do. That was, to fix the situation that the carriers have right now, which is that they are not able to be profitable.

They are not able to be profitable in part due to the complexity of the current situation, which has two parts. The first part is that there is [generally] a network for each application type, where because the network is built specifically for this application, you are able to get the reliability, security and quality that the application needs.When you do that, you are doing two things which are bad. Number one is that you've damaged the any-to-any ability of networks by carving things into silos that don't communicate with each other. You rob the value that you can otherwise provide. Number two, because the technologies and the skill sets needed to run these diverse networks are different, you don't have economies of scale anymore. The same carrier might have to employ three or four different sets of people, one for each network type.

Advanced IP Pipeline: You're talking about Frame Relay, ATM.

Sindhu: I'm talking about the TDM network, the Frame Relay network, the ATM network, the cable TV network, the mobile network. For one application type, you have a specialized network. That's one situation.

The second situation is the Internet, which started out with the goal of being this ubiquitous network infrastructure that could support lots and lots of applications. If you look at the situation today, there are applications that are not so demanding [of the network], and those that are very, very demanding. The Internet is able to serve the 'best effort' class, which includes e-mail, web browsing, and overnight backup. But it doesn't do a very good job at all of serving those [that are more demanding]. With those applications, you have jitter-sensitive, very high reliability requirements; if the call goes away for more than a few seconds, or even a second, you conclude that something bad has happened.

Generally, more demanding apps have higher reliability, higher quality, and higher security requirements. The Internet doesn't do a very good job of doing that.What the Infranet is, is two things: A vision for where we want to get to; and more importantly, it gives you a path from where we are today. It's very easy to put forth a vision, then stop there. That leaves a harder and open question of how do we get there.

The main reason current situation is untenable is that neither the half-a-dozen to dozen different technologies, nor the best-effort Internet, solves the problem of making carriers money. The first, because it's too expensive; the second, because it doesn't support the applications where you can make money. So you have a quandry, and what do you do?

The first question we asked ourselves, we believe is already answered. [That question is,] is it possible to build a single infrastructure, based on IP, that can support a very broad range of communications applications? The answer to that question is yes. There's been a debate in the industry for the longest time, [but] I think that debate is largely now settled. People used to disbelieve whether a switched infrastructure could support broad array of applications. The answer is, yes it can.

Now -- once you've answered that in the affirmative, whether IP and MPLS is the right technology, and you say yes, they are the right underlying technologies, then you ask, what is the gap analysis between where the Internet is today, and where the Infranet needs to be? No question about the fact that the base technologies, IP and MPLS, are the right ones, so what's missing?

There's really three categories of things that are missing:Number one -- a business model that carriers can have that can make them profitable. This is really a joint industry effort -- no vendor alone, no carrier alone can do this.

Number two -- there are certain standards that need to be put in place for an effort like the Infranet to succeed. If, for example, the way in which client applications interact with the network, and the way in which carriers interact with other carriers is different from one carrier to another, and different from one client to another, you will have a mess. A very, very, complex mess. Things like this can only be addressed by standards -- worldwide, open universal standards.

What we would like to do is create two standards affiliated with the Infranet. One is called the client-to-network interface, where an application can tell the network what attributes it wants the network to satisfy, such as what level of security and quality it needs, and how long it needs them for. It passes this information to the network, and the network can then, in return, say whether or not it can satisfy the request.

The second is a carrier to carrier interface. Inside the network, [there is] more than one carrier. The problem there is that, for one request to get from its source to its destination it will probably pass through multiple carriers. If I'm not able to express the communication's needs from one carrier to another, I lose the ability to provide the high quality, reliability and security it needs.

Advanced IP Pipeline: And then there's the billing question.Sindhu: Carriers do not have an incentive to carry other peoples' traffic, unless they are paid for it. The billing question has to be settled in the context of security, reliability and quality. It's still going to be reasonable to pay more for increased reliability or quality or security. And to pay less if you don't have these.

Advanced IP Pipeline: Right now, it seems like a mess for some carrier-to-carrier communications.

Sindhu: In the communications industry, one thing which is remarkable is that one even has to explain that standards are exceedingly important. The objective of the industry is to communicate. In order for two people to communicate, you need a standard -- otherwise they don't communicate. This is so self-evident as to be belaboring the point to even say it. The fact is, there are these two standards we really need.

So we clearly need the cooperation of other vendors, other carriers, and we're trying to get as much mass behind that as possible.

Now there's a delicate balance here -- as we all know, work doesn't get done in committees with 30 or 40 people. You have to form small workgroups that can take off these bite-size chunks and actually address them.The third piece that's missing -- if you look at the network as it is today, a significant proportion of the equipment doesn't have some of the mechanisms that would be necessary to support higher quality, reliability and availability. You have to ask, how many compute cycles for packet processing are available, per packet, as the packet slides by?

How much of this processing is available completely defines whether I can provide the right levels of security, quality and reliability. In security, for example, the problem in the network is to figure out whether the packets I'm transporting are good ones or bad ones. In order to do that, I need to look inside packets.

Another example: If I want to support high quality, in terms of low level of jitter, I need to do differential queueing. In order to do that, I need to look inside the contents of the packet, to figure out what class it belongs to, because I can't trust the class painted by the end user -- because the end user will paint everything 'high priority.'

What we believe is that a lot of equipment installed currently was designed in the 'best effort' era, and it doesn't have the appropriate mechanisms to support these additional features. Unless the equipment over time...

Advanced IP Pipeline: You're talking about IP equipment.Sindhu: This question doesn't arise with TDM equipment; TDM equipment, in terms of QoS, is the perfect thing. A bit gets from here to there without jitter, or very little jitter.

Advanced IP Pipeline: Making it happen the hard part. When does the Infranet...

Sindhu: I was about to explain where we are. Between that [February] meeting and now, we have had one-on-one meetings with a lot of people. Mostly carriers, mostly customers. By now, there are about two dozen people who have expressed reasonably strong desire to participate. On June 21, we're having the first formal meeting of the Infranet Council [at Supercomm in Chicago].

At Cannes, people at the meeting felt it was important to come up with a reference model, outlining the scope of the work, etc. Let's get that defined. We now have that kind of defined, and one of the objectives of the meeting [next week] is to get a consensus on that. To make sure we have the right working model and architecture.

There is work going on which is similar, with some overlap, in various bodies, like the ITU, and the IETF. We do not want to ignore this work. At the same time, we believe that no one has the focus we have. If the problem was solved, we would not need to do anything. We want to build bridges to the appropriate organizations, to make sure that on both sides people are aware we're doing this work, and to the extent we can leverage work that's already done, we'd be happy to do that.Then we need to form working groups, each of which is focused on a piece of the problem, and to make rapid progress toward something real.

Advanced IP Pipeline: Have you talked to Cisco yet, in a formal way?

Sindhu: We have approached Cisco, and we're not going to say anything more than that.

Advanced IP Pipeline: Can this happen without Cisco? Can it start to happen without Cisco?

Sindhu: I think that there is a lot of interest by the industry, specifically by the carriers. The main pain point is that they would like to be more profitable. This Infranet Initiative addresses that head-on. That is the reason for its existence.Advanced IP Pipeline: What are the most important things carriers are looking for from equipment providers?

Pradeep Sindhu: They are looking for networks that can run a much broader range of applications. They are looking for ways to get cost efficiencies from their network. And they're looking for help, to get from where they are today to one single IP-based network for all applications. In this sense, the Infranet Inititive is precisely tailored [for carriers].

Advanced IP Pipeline: Are these new features table stakes, things like integrated security?

Sindhu: They are table stakes. You cannot even begin to play. It's a usual thing in any industry, to reach a level where things are taken for granted. They're not taken for granted at the moment, since we still have the best effort Internet. We are trying to change that.

Advanced IP Pipeline: Are there any components not integrated now, that need to be? There are a lot of appliance-type technologies in the networking space. Will Juniper need to integrate any of these?Sindhu: From time to time you'll see clever ideas that have not been incorporated into the base infrastructure [of high-end equipment]. Sometimes the idea will cross the threshold of being interesting enough, and you will need to incorporate these set of functions or features directly into the infrastructure. I suspect this will happen more at the edge of the network than the core, because the edge is where most application-level knowledge is.

Advanced IP Pipeline: Does Juniper play any kind of a role, maybe via the Infranet, to helping settle problems between carriers, like access-fee battles?

Sindhu: We see our role as a supplier to the carriers. We also see our role as a catalyst for ideas. When we saw a state of confusion after the downturn, we felt it was our responsibility to come forward and make a statement, and gather people around this idea, so we could lead this industry in the right direction. The positive response from people seems to support that.

I don't know that it's Juniper's role to get involved in the legal issues of how telecom regulation should be done. I'd just say that anything that promotes innovation in this area is good.

Advanced IP Pipeline: Does Juniper do any lobbying?Sindhu: I don't think we have a direct effort.

Advanced IP Pipeline: I think everyone may agree the Infranet Initiative is a good idea. The problem with these kinds of efforts always seems to be that when one company is too visible a leader, it's hard for some companies to get behind the idea. Java, for instance, still suffers from being seen as mainly a Sun product. Are there any of these possible roadblocks to adoption of the Infranet?

Sindhu: Let me state very clearly : No. 1, our goal is to surface the issues, identify the gaps that are there, and be a catalyst for filling the gaps in a particular way. We insist that they are an open standard. Juniper has never implemented a closed standard in its entire history. We don't believe in that. We don't believe in that precisely because of the very nature of communications. Our customers all agree that they want vendors to support open standards.

For something like this, to have one mediocre standard is far better than three or four 'perfect' standards. The reason is subtle but simple: If I have two standards, I'm taking an entire marketplace and dividing it up into two pies. And the size of each pie does not have critical mass, and this is a bad thing. When I have a multiplicity of standards, it's even worse. Look at the problem with [power] plugs -- a stupid daily example. If I take a plug from here to Europe, I can't use it.

When the very essence of technology is to communicate, you need to plug things together. Having multiple standards mean everybody loses. In the end everybody pays, and the market is smaller than it could be by far.Advanced IP Pipeline: It's an ambitious undertaking.

Sindhu: It is ambitious, but we believe it is important. We're not so much into making statements. At the end, we want it to make a difference.

Advanced IP Pipeline: Technologically, how do you feel about Juniper these days?

Sindhu: Juniper today is better positioned than it has ever been. We have a unique position in the industry. I feel very good about the people we have, and the industry looks like it is recovering.

Advanced IP Pipeline: Is your product mix enough for what carriers want?Sindhu: We always have stuff in the pipeline. We take great pride in being close to the customer, and having the equipment customers need, when they need it.

Advanced IP Pipeline: At Cisco's CRS-1 announcement, they were asked if it's good to have Juniper as a competitor. Is it good for Juniper to have a competitor like Cisco?

Sindhu: I think competition is always good. Everyone eventually benefits. Competitors benefit because they are more efficient, and they advance their products. And then customers benefit as well.

Stay informed! Sign up to get expert advice and insight delivered direct to your inbox

You May Also Like

More Insights