Modern Infrastructure

The problem with private cloud


Data center SDN still in its infancy, says expert

Data center SDN means speedier access to business assets, but there's a lot to consider, such as cost, before taking the plunge.

Vendors say that software-defined networking can solve many of the issues that networks have traditionally posed for virtualized environments. Data center SDN can help shift environments closer to the cloud by bringing the network to parity with other aspects of virtualized infrastructure.

Bob Muglia, executive vice president of software and solutions at Juniper Networks, talked with Modern Infrastructure about the emergence of SDN, what it means for enterprise data centers, and why IT managers should care about this still-evolving SDN technology.

Prior to joining Juniper, Muglia spent more than two decades at Microsoft. Most recently he was president of the Server and Tools division and worked on some of the company’s key virtualization- and cloud-based technologies. As Muglia noted, “When we were building System Center and Hyper-V, and when we were building Azure, . . . the lack of dynamic capabilities in the network was a significant bottleneck.”

Modern Infrastructure: What are the potential benefits of SDN for an enterprise-class data center?

Bob Muglia: We have watched companies move from an environment with physical infrastructure to one where they have a virtualized infrastructure. They are now looking for more of the dynamic attributes they can derive from a cloud-based infrastructure.

The difference [with a cloud-based infrastructure] is the need for the network, the whole system to respond. In a virtual infrastructure, when a business unit wants to put up an application, they send out an email to the IT department, who says, “Great. We’ll get this up within the virtual system in a couple of days.”

Do you run a private cloud?

In a cloud-based infrastructure they go to a portal, they talk about the resources required, and 10 minutes later the system has spun up all those resources that are available for use.

And in that world the underlying elements of the infrastructure -- the compute, the storage, the network -- all have to dynamically respond to the changes required. And what we have seen is, the compute infrastructure first, then storage infrastructure, have been quite responsive to the needs of the business, but the network has not. So that is the business problem that SDN is focused on solving: to make the network dynamically respond and to meet the needs of the business.

Why has traditional, physical networking been a bottleneck?

There are a lot of elements in the network, a lot of devices -- way more than there are storage devices, for example. Basically each device is configured independently and the configuration data is stored and mastered within that device. So there really is no centralized point of control, no standardized way to automate that process. That has been the fundamental issue.

Are there benefits in terms of cost and performance?

Yes, the biggest benefit of cost will be the operational cost that is reduced and the people cost associated with managing the system. If you look at any IT shop, the dominant cost is the people side. [So] you will see a reduction of cost because there is a lot of manual complexity associated with running the network.

Will Desktop as a Service drive more virtual desktop adoption?Are there factors that might hold back acceptance of data center SDN, such as in the area of standards?

The biggest issue that will hold back adoption is the state of maturity within the industry around this. To say it is immature would be an understatement. This is early, early, early days of it. Less than 1% of all data center systems are SDN-enabled.

Over time, as the industry goes through its typical maturation process, standards will emerge. [And] while OpenFlow is an interesting standard, there are many others that need to emerge as well.

One is XMPP. We think there are standards existing today that can be used effectively, standards for data path in particular such as MPLS over GRE or VXLAN. Those standards are usable as-is.

At this point, there isn’t any one protocol where we sit back and say, “This must be the answer.” The issue isn’t so much the protocols as the architecture of the solution. The more macro point is that we do need to see broad-based adoption of some standards.

For those getting ready to implement their first SDN initiative, what sorts of things should they think about?

  1. Centralizing management. This is a very achievable thing to do in a shorter period of time and generates significant benefit in operational cost savings, to move away from a distributed, current, typical approach of managing each individual network element as an independent unit, such as [Juniper’s] Network Director product.
  2. Evaluating network services. Today these services are chained together in a physical way; I have an application delivery controller box, I have a static firewall box—different boxes that are connected together. [Shops can look at] which of those services could be virtualized and pull those things out, and begin working with virtualized services.
  3. Centralized controller architectures. This is where many of the benefits of SDN are fully realized. But it is still fairly early for this.
  4. Optimize the usage of network and security hardware to deliver high performance.

Do you think there are certain types of shops where SDN is not the answer?

Unless you are contemplating installing a private cloud, SDN doesn’t make sense. There is some minimum threshold of servers that are needed to justify a private cloud. Truthfully, no one knows what that is today. I can easily tell you if you have 500 or 1,000 servers, I think it makes sense. If you have 50 servers, is it justified? Maybe. Fifty servers colocated in a data center, well, you know, there is a good chance it could benefit you. If you have 50 servers located across 10 branch offices nationwide, then [that’s] probably not enough.

Let us know what you think. Write to us at

Article 8 of 14

Dig Deeper on Data center capacity planning

Join the conversation

1 comment

Send me notifications when other members comment.

Please create a username to comment.

An anti-SDN position is to-be-expected from Juniper. They along with Cisco and Alcatel-Lucent have the most to lose asnetworks move in this direction.

Get More Modern Infrastructure

Access to all of our back issues View All