February 26, 2013

262 Views

Network Virtualization

Networking, one of the three vital pieces of physical IT Infrastructure apart from servers and storage, is  at a tipping point today. The cross currents of virtualization and cloud computing have already triggered beneficial changes in servers and storage space, and now networking, too, must fall in line in the march  toward an efficient converged infrastructure.

What are the Pain Points?

Virtualization is a longstanding IT paradigm which decouples the logical from the physical Infrastructure. Some of its major attractions are greater efficiencies, speedy provisioning and above all, demonstrable cost savings. Over the last decade, enterprise data centers have started investing heavily in server and storage virtualization to reap these benefits. However, we cannot say the same about the networks - it remains pretty much hard wired even today, consisting of monolithic stacks proprietary to a few vendors. We know that now more than ever, applications are being consolidated and deployed in virtualized or multi-tenant cloud environments.  While thanks to server and storage virtualization in the datacenters, these resources can be obtained quickly, but a network roll-out can still take a week.

The primary reason is that the network’s control logic remains physically embedded in the end points and devices. This hard wired and static paradigm severely restricts a network’s programmability potential in the cloud and in virtualized setups where it is normal to manage virtual servers and storage resource pools using APIs. Thus, networks clearly need abstraction mechanisms which can decouple the logic from the underlying physical pipes and wires so it can become more aligned with the applications they are there to support.

Also, traditional networking is just not ready to take on the gargantuan demands of scalability required by cloud computing on a sustainable basis. Network topology is currently configured based on static hosts, first into virtual LANs (vLANs)/Layer2 (L2) IP subnets, and these are then demarcated by Layer 3 (L3) physical IP networks. This practice has proved to be restrictive for efficient scaling of large scale multi-tenant cloud infrastructure. In these environments, inter-host workload mobility is frequently required to optimize the usage of the underlying virtual servers and support other complex use cases. Effectively, this translates to migrating VMs between vLANs or over L3 networks. In the modern networks, a vLAN is restricted to 200-300 VMs and it can connect a handful of racks, which is as far as the VM migrations work fine. However, VM migration across the L3 network is a big issue today. In short, shifting of a dynamic virtualized workload around L3 boundaries is becoming more of a standard requirement in dire need of sustainable, long-term solutions.

What is the way out?

The situation called for a breakthrough, and thankfully, there are multiple mechanisms and approaches evolving to deal with it. Notable among them are the concepts of overlay networking and software defined networking. Collectively, they will be driving the networks on the lines of virtualization. It is an idea whose time has come for networks. Let’s look at these concepts a little more closely.

  • Overlay networking: Overlay networking is about creating a virtual network to which the traffic from each VM is mapped. The traffic is then routed seamlessly over the current physical infrastructure across L3 boundaries. Behind the scenes, the virtual networks to which the VMs are mapped are actually L2 overlay schemes over an L3 network. This facilitates seamless virtual L2 connections over two or more physical L3 networks. For the virtual workloads, Virtual L2 connections create the appearance of being on same L2 subnet. Cloud style scalability is thus enabled by facilitating VM mobility across L3 networks in this manner.

Although the overlay network model is gaining popularity, there are concerns regarding its ability to deal with the constraints imposed by the underlying network infrastructure. Network performance can degrade with this model in heavy traffic scenarios such as a hybrid cloud model where many customers converge on same physical infrastructure. We can expect the overlay networking model will perform better with a high-bandwidth L3 network. Complex workloads with high-bandwidth requirements to multiple servers can get starved of network connections in democratized, multi-tenant cloud environments.

  • Software defined networking: In this model, programmable abstractions are used to decouple the control logic from the physical network infrastructure so the application can tell the network what to do. In this model, control logic and policies can be defined, and later modified, using a centralized controller or hypervisor making the networks more dynamic, flexible and easier to administer. This has come to be known as “software defined networking,” or SDN. While SDN also facilitates VM mobility across L3 network boundaries, the promise of SDN is more expansive. SDN’s power lies in its ability to decouple applications from the underlying networking infrastructure. Not only can this herald an era of better flexibility, cost effectiveness, innovation, and control in managing the underlying network fabric, but it can also potentially trigger unprecedented levels of alignment between applications and networks.

SDN can be used to build effective network automation tools for challenging use cases in the arena of multi-tenant cloud environments such as V2C migration and public or hybrid cloud network configuration for enterprise users. Speedy application deployment is another benefit that is being posited. With the option to define network requirements within the application packages, it would be possible to deploy applications anywhere so that they work with any underlying hardware. Network management costs, too, are expected to go down with SDN. SDN is still in its infancy, and various approaches for the actual implementation of controllers are still being developed and refined by various vendors. The potential for innovation it puts in the hands of programmers and the business agility it could facilitate are some of the reasons why the SDN space has become a hotbed of fervent acquisitions and active research in the IT industry today.

What’s around the corner?

As virtualization is now poised to permeate to networking too, managing the security of end-to-end virtualized infrastructure will become more important than ever. There will be a pressing need for network virtualization tools to manage network resources for applications, and they, too, must evolve at the same pace. An overlay network and SDN can be competitive in regard to their features and use cases. Market acceptability will most certainly be a deciding factor. Vendors such as Cisco, Emulex and VMware are backing the overlay networking models, and the technology is looking promising. Due to SDN’s ability to build a more programmable network, new modular products will evolve in place of today’s monolithic vertically-integrated networking stacks. The existing networking business models can get disrupted as physical network elements get commoditized. By 2016, IDC expects the SDN market to grow to $3.7 billion. Apart from the leading networking vendors, startups and many software and virtualization vendors are jumping onto the SDN bandwagon. Recently, Cisco announced that cloud and SDN are its future focus areas. Citrix has already introduced an SDN-aligned “app-aware” variant of its flagship NetScaler networking platform. IBM, Microsoft, Dell, Juniper and HP, too, are backing up SDN strongly. Nicira, a startup specializing in software-defined networking, was acquired by VMware last year. Intel and many others are funding Big Switch, a startup in the open software defined networking space. With such a vote of confidence being placed on the upcoming innovations by one and all, we can be sure of the fact that network virtualization is the next step for enterprise data centers and cloud environments.

References:

  • Simplify Hybrid Cloud Deployments, by Brandon Hoff, August 8, 2012, Cloud Computing Journal, SYS-CON Media

  • Overlay Networking: vLans for the Cloud, by Arthur Cole, July 17, 2012, IT Business Edge

  • Software Defined Networking: A Dell Point of View (whitepaper), October 2012,  Dell website 

  • SDN Market Forecast at $3.7 Billion by 2016, By Sean Michael Kerner, December 20, 2012, Enterprise Networking Planet, IT Business Edge

  • Software-Defined Networking Gains Intel Capital Investment, February 11, 2013, by CJ Arlotta, MSP Mentor.net, Nine Lives Media Inc.

  • Introduction to OpenFlow and Software-Defined Networking, By  OpenNetSummit, Youtube

  • Software-Defined Data Center, By TheCloudcastNET, Youtube

  • An attempt to motivate and clarify Software-Defined Networking (SDN), By EricssonTechnology, Youtube