C-Change 2013: Busting the myths of SDN

BANGALORE, INDIA: Sajan Paul, director-Systems Engineering, Juniper Networks, in an interaction with Deepa Damodaran of CIOL, on the sidelines of C-Change 2013. Excerpts:.

CIOL: Can you tell us what are the networking challenges that CIOs are facing in cloud and BYOD?

Sajan Paul: Cloud and mobile are the two major trends. On one side you have this repository of information, which is cloud, and on the other there are these devices from the consumer side and then there is this distance between cloud and device, which is the network.

Today the network is not the same that used to be three years back. It has changed completely. Types of applications and behaviour of these applications within cloud have changed.

Compute power is doubling every eighteen months (Moore's Law) and storage is also doubling by each day. however, we do not see the same happening in networking. So far, only the speeds and feeds are changing, whereas the whole architecture of the networks have remained the same.

So the challenge is how to transform this piece of network so that the consumer side and cloud side come together.

CIOL: Juniper is very strong in the services segment, why did it get into the enterprise segment?

Sajan Paul: Our genesis was largely around service provider. When we invented the fastest router in the mid nineties, the internet world was on the verge of collapse because the nature of network was largely like an enterprise network.

During mid-2000, in 2007, we started our journey into enterprise segment because the boundary between a service provider and enterprise was narrowing and some of the features that service providers wanted, even enterprises wanted. For example, availability, quality of service, traffic engineering, etc. Since we have already invented in the service provider domain, we thought why cannot we use the same innovation in the enterprise domain as well? That is why we began our journey. Our focus is on routing, switching and security.

We are becoming an enterprise company as well, as much as we are a service provider one. Our revenue share is almost 50-50 among these segments.

CIOL: How has Juniper adopted SDN and how is it transforming your business?

Sajan Paul: Juniper is not a traditional company. We are born into the IP world, unlike our competitors who have a legacy of TDM or old generation network. We did not have any baggage to carry, so ground up our designs are IP centric. In terms of architecture, we did two things, we centralized what can be centralized and we distributed what must be distributed because speed and performance come by distribution information and intelligence comes by centralization.

That gives us an edge on the SDN front, which is about four layers - control plane, management plane, services plane and forwarding plane. SDN is about centralising =g control and management planes so that software gets the control. That means your business applications can interact with these two planes. The forwarding plane has to be as close as possible to user because that is where data gets switched and distributed.

So we are innovation on the control plane which takes control of the forwarding plane and we centralize that Juniper QFabric has an SDN architecture. You take the control plane out and distribute the forwarding plane wherever you want, and get 4Gig or 100Gig performance.

CIOL: Does SDN mean hardware will no more be important and that hardware manufacturers will perish?

Sajan Paul: There are several such myths associated with SDN. They are:

1. Hardware is not important

We believe hardware value will go up. Today, the challenges is that most of the vendors are building network in a monolithic way. That means all the four planes that we talked about are built into one operating system, which makes them very difficult to separate. And, in order to separate them you need a new operating system.

Juniper's operating system has already been segregated from these planes. Junos vApp Egnine is about taking out control and management planes out of the box, run it on an x86 in the cloud. World's fastest router, T Series, MX and PTX, run as boot force hardwares forwarding engines. The entire control plane engine is sitting in the cloud in some x86 server. There are over 500 application developers, who build applications on top of it.

What made an iPad so popular? It is just a collection of silicon and a design house. What made it important is the number of application developers who make it available to users. Network did not evolve in this manner. We do not have a single network application which can be ported to routers.

SDN will be successful if we make network as a platform and allow application vendors to write applications and execute it on top of the network.

2. SDN is all about software and no hardware

Many believe that networking vendors such as Cisco, Juniper, brocade will go out of business because in SDN you can pick up hardware from any store and then run OpenFlow and it works.

When you have 80 per cent market-share, do I want to jeopardize that by saying you buy anybody's hardware or controller? Obviously that is a revenue loss. That is the thought process of many traditional companies.

This is a myth. The challenge is we invest 25 per cent of our money in R&D into silicon. SDN gives the intelligence to the forwarding plane, but what if the packets move at snails pace. So you need packet forwarding at the speed of the light. While software does this, we need to innovate on the hardware and on custom silicon so that you get the best of both the worlds.

3. SDN is all about data centre

SDN may begin in data centre, however, will not stop there. We are looking at a concept called service chaining. From a service provider's point of view, when a subscriber subscribes for a service, how does the network gets programmed for that service, which maybe a lease line service, or layer 3 VPN service or layer 2 VPN service, or IPv6 migration service etc. So how does this service chain interact with each other.

This is a big area of focus today. We are building SDN to control that piece. So, the moment a user comes in, his/her complete service chain is formed. We are also building SDN layer on top of security portfolio so that when a threat is identified in any part of the network the SDN control plane can interact with the elements and block it. So SDN is beyond data centre.

4. SDN is very far away

We do not think so. Is it now? It is not. However, it will happen in another one to two years time. Till then, you will see small deployments and advancements in small scale. However, for hardware to become agnostic, it is little farther.

5. SDN Software Licensing

We do a lot of services on our platform. All these services need to change with SDN. But how do we buy licences for that? Today licence is attached to a hardware. So if a router goes out of business you take it out of network, your licence also goes so licence become perishable.

We are looking at a perpetual licence. It will not be linked to the appliance. We will track it around five metrics from a licensing point of view. We will be segregating it from the hardware and track it based on consumption model.

CIOL: Why do you think SDN standards have to evolve beyond OpnFlow protocol?

Sajan Paul: There are protocols that are evolving, such as OpenFlow and BGP, Alto etc. OpenFlow is a great protocol when it comes to data centre networking where everything is in your control. However, when you go down to a large network, such as that of a service provider's, which has several large PoPs, you need a protocol which also bring in real time topology.

So we are working on certain advanced protocols which will enhance the advantage of SDN.

SDN is a fashion statement today and there are several start-ups around it. A lot of these newcomers do not have a networking background. Networking protocols that have evolved over a decade do not have any networking background.

If you see recent Google's announcement regarding SDN adoption in data centre, it is largely talking about data centre applications such as how a replication data is being routed through the most optimal point or how a link with lesser quality of service is made to better utilise capacity of your network. These are applications that they have written into a normal commodity x86 server and allow that to communicate.

Here OpenFlow works very well, however, what it does not do is to go and figure out a dynamic topology of a network. A large service provide will have hundreds and thousands of network elements. They all communicate BGP, OSPF, MPLS protocols and OpenFlow is not designed to communicate to these protocols. OpenFlow is very important and has a major role to play, however, SDN journey does not stop there. We need to innovate a lot.

CIOL: Is Juniper a bit slow on the SDN front because companies such as HP, BigSwitch are already talking about rolling out solutions?

Sajan Paul: Not exactly. There are certain newcomers and what they talk about is largely from the control part. What they have is an Opensource programme, which they can port it into a controller appliance. They may also build hardware. However, then, will a BigSwitch work with a Nicira device? Very unlikely.

Inter-operate is a big element that is yet to be established. So are we looking at BigSwitch giving a switch and controller together and then building again a monolithic network? This is what is going to happen. Juniper can do this today, however, we are looking beyond that. We do not want to go into the part of launching a controller, which we can do it in no time since it is possible on our chip Trio, and say we are ready for SDN. That is not the strategy.

CIOL: Is being a pure networking player a disadvantage because storage company EMC is building its networking capability, and networking company Cisco is building its server/storage capability?

CIOL: Neither compute vendors nor storage vendors are innovating on network side and they do not have the capability to do so.

The business model that you have to look at it out of $10 dollars spent on data centre, only $2 is spent on network, the rest $8 is spent on compute and storage, power cooling. Their business model is why should not we go after the remaining $8 dollars? It may not be about building the best in class or technology innovation.

Compute power is going over the roof, and storage capacity is increasing by day. Efficiency of data centre works on two principles. Centralisation - the more you centralize the better the data centre is. the second is Pooling effect - how many resources can you pool and do parallel computing. Thus, you get square root of number of pools worth of efficiency in that data centre. That is why data centre consolidation and cloud become more efficient.

However, for these two to work together you need a network at the speed of light because the VMs have to move statistically across network, otherwise you will never capitalize on the ability of virtualization. That is why innovation on network is required.