White-Paper-Next Generation Access Network Architecture...processor technology available to central controllers. It was the right tool for the problem at that time. Controllers eased

Next Generation Access Network Architecture Protocols are Free

White Paper

2 Copyright ©2012, Aerohive Networks, Inc.

Table of Contents

Introduction .......................................................................................................................................................... 3

The Dwindling De Facto Standard ..................................................................................................................... 4

Why Distributed? .................................................................................................................................................. 7

Cooperative Control ......................................................................................................................................... 10

Economic Deficiencies of a Controller-based WLAN Architecture ............................................................. 10

Economic Benefits of a Cooperative Control WLAN Architecture .............................................................. 12

Deployment Examples ...................................................................................................................................... 15


Introduction There are three "planes" of operation in networking infrastructure:

§ Management: The management plane is always centralized and encompasses functions like pushing configuration and firmware updates to infrastructure devices, pulling statistical data for producing reports from those devices, and license management.

§ Control: The control plane is the set of real-time operations within the infrastructure, such as controlling connections, disseminating connectivity information, and calculating optimal path. A shared control plane in any infrastructure system can be achieved in either of two ways: centralized or distributed. In both switching and routing, the control plane is distributed, operated by protocols (e.g. spanning tree, OSPF) between intelligent devices.

§ Data: The data plane is the data frame/packet forwarding path and can be centralized or distributed. In both switching and routing, the data plane is distributed, allowing for best-path forwarding of data flows, which offers the most reliability and scale.

Prior to the introduction of the WLAN controller (a hardware platform that centralizes WLAN control plane, hence the name “controller”), circa 2003, autonomous APs had no method of sharing the control plane among themselves, so the industry chose to go with a centralized control model as the de facto standard, rather than the traditional distributed model. This choice was due:

§ The price of access point components (CPU, RAM, TPM, etc.) required for a distributed model with a shared control plane in 2003 was cost prohibitive.1

§ The convenience-focused, non mission-critical nature of Wi-Fi infrastructure at that time. 802.11a/g networks were slow, very few clients were available (remember the smartphone explosion wouldn’t occur for another 4-5 years), pervasive coverage was rarely required, and Ethernet was readily available.

The controller served the purpose of a central “brain”, and controller-based "thin" APs served only as radios/antennas, separated from the brain by an Ethernet cable. Ten years ago, these first generation of controller-based architectures supported only centralized data forwarding because the relatively low volume of data at the time (802.11b/g only supported data rates of 54Mbps) and the fact that the WLAN was only for convenience and coverage was limited common areas and conference rooms and therefore the processing power required could easily be handled by the then-current processor technology available to central controllers. It was the right tool for the problem at that time. Controllers eased the management and security headaches that non-pervasive network comprised of autonomous APs would cause and the centralized

1 Moore’s Law would prove to be the saving grace. 4-5 years later, the cost needed to support a distributed control plane would be a fraction of what it was in 2003.


control and forwarding allowed the controller to process data for security as well (since that thin-APs did not serve that function either).

Today, three unstoppable forces have combined to make this centralized model a severe architectural limitation:

§ Faster speed: Wi-Fi can now achieve data rates of 450 Mbps per client—833 percent faster than in 2003.

§ Distributed access: Access to enterprise applications is increasingly distributed among end devices, users, and locations.

§ Powerful smart devices: Smart devices sport dual-core, 1Ghz processors with graphics co-processors and increasingly use the latest video and voice applications.

Many WLAN products that were designed for wireless networking problems of 10 years ago attempt to reto-fit their controller architecture to deal with a more modern day network by placing some features in the AP and others in the controller (often called the “split MAC”), supporting partially-distributed data forwarding and forcing a choice between scaling or feature support. They have even gone so far as to use modern virtualization technology to embed the “controller” into a virtualized server (a la VMware) or embed the controller into an AP (essentially using the APs processor resources as a mini-controller). The fact remains that control is centralized and that is what ultimately limits scalability and increases operational complexity.

Regardless, the fact remains that the combination of massive growth in wireless and mobile usage and increasingly distributed nature of work simply breaks down the centralized model.

In 2007, Aerohive re-introduced what every router and switch network already used as a defacto standard – fully distributed control plane leveraging free protocols with a centralized network management system, eliminating the need for a centralized controller. This architecture, called Cooperative Control Architecture, allows every device to share in the processing of wireless data.

Fully distributing control with the Cooperative Control Architecture achieves three main benefits:

§ Cost Savings: By replacing controller hardware, software, and licensing with free control protocols dramatic cost savings can be realized without losing functionality.

§ Operational Simplicity: Using a distributed control plane allows the WLAN devices to self-organize and integrate directly into the access architecture and enforce security policy before WLAN traffic ever traverses the wired LAN.

§ Scalability and Flexibility: With every access point or networking device participating in the processing of data, much like a grid computer, the network can provide full functionality to any network regardless of size. Starting with the first Cooperative Control device (AP or Router), the network has the full functionality of an enterprise WLAN with centralized management. Every device added to the network increases not only the coverage but the total compute capacity of the network.


Centralized Control Solves Yesterday’s Problem Yesterday's de facto standard architecture is commonly referred to as the, “controller-based” architecture. It involves one or more controllers and controller-based (lightweight, thin) access points (APs). The controller-based architecture was created to solve manageability, mobility (as opposed to portability), control-plane inadequacies, and high operational expenditure (OPEX) problems that were prevalent in autonomous (fat, thick, standalone) AP implementations.

Figure 1 - Controller-based Architecture (Controllers and Thin APs)

While the controller-based architecture solved some manageability, mobility, and control-plane problems, the introduction of controllers created new problems, such as:

§ Higher capital expenditures (CAPEX) due to increased hardware and licensing, especially when redundancy is required

§ Bandwidth bottlenecks, especially with 802.11n deployments

§ Added latency and jitter due to traffic u-turns at controllers, especially when filtering at the controller

Creating the controller-based architecture was a divergent path to solve some Wi-Fi problems of the day (circa 2003), but Aerohive went back to the starting point (2007) and improved upon the original solution (autonomous APs). We call it Cooperative Control. Instead of starting with an architecture where a costly, centralized control platform is needed, Aerohive followed the original intent of the 802.11 standard designers more closely and brought the technology to maturity. Now inter-AP protocols execute the shared control plane functions performed by centralized controllers, but without:

§ The high costs of primary, redundant, and branch controllers

§ Non-linear scaling costs of controllers

§ The high costs of AP and feature licensing on controllers

§ The high costs of rack space, power, cooling, and support for controllers


§ Re-architecting the network to provide for an overlay Wi-Fi infrastructure

§ Single points of failure without add-on costs

§ Multiple high-capacity (10G) core Ethernet switch ports

§ A long learning curve for administrators and engineers

§ Traffic bottlenecks and latency

§ Management & deployment complexity of non-essential components

§ Security holes created by centralizing security features

The key takeaway here is that inter-AP protocols are free, but controllers (and all of their associated costs) are not. In a market where all enterprise-class APs cost roughly the same, removing the controller hardware and feature licensing from the equation results in an immediate and extremely significant CAPEX decrease. For networks where controller redundancy is required, or there are many locations (e.g. branches) where shared control plane features are necessary, the cost savings can be shocking.

When Ethernet switches run intelligent switching protocols and routers run intelligent routing protocols, why should access points be dumb? APs are distributed devices doing distributed computing, and they should run protocols that allow them to coordinate distributed data flow. Aerohive’s Cooperative Control protocol suite does the controller’s work, which eliminates the need to redesign your network in order to introduce a controller-based overlay infrastructure. That’s one less thing to manage, deploy, configure, and pay for.


Why Distributed? Let’s look at why network infrastructure has always been “distributed” in general.

The most simple network infrastructure possible is just a single switch-router, as shown below in Figure 2. It’s an all-in-one box - it does everything. However, the moment you need more ports or physical separation of ports, you’re forced into a situation where you have to choose between centralized or distributed intelligence. How will the two or more boxes be coordinated (shared control plane)?

Figure 2 - Switch/Router, Simple Network

For a switched network, as we break it apart, we need to introduce the Spanning Tree protocol between the switches to ensure a loop-free topology. Having multiple switches using a protocol gives us a distributed, intelligent, scalable, and resilient, network.

Figure 3 - Switched Network with Spanning Tree Protocol

For a routed network, we need to propagate routing information (shared control plane). Breaking apart a single router, the simplest possible network, will require a routing protocol such as OSPF or BGP. Just as with the switched network, all forwarding will be distributed, the best path will be selected, and backup routes will be available for added reliability.


Figure 4 - Routed Network with OSPF Protocol

Controllers force trade-offs due to an architecture that is ill-suited for today’s Wi-Fi challenges, such as high-throughput, high-density, high reliability, pervasive coverage at low cost, and deployment flexibility.

Figure 5 - Controller-based Architecture, Trade-offs Required

Aerohive’s Cooperative Control works in the same manner as the Internet and your routed and switched networks: enabling distributed intelligence, fully-distributed forwarding, and all policy applied at the edge of the network through control plane protocols. The network architecture is as simple as plugging in APs to Ethernet ports (Portal APs) and where needed, powering APs with AC power to have them connect via mesh (Mesh APs) to Portal APs. Voila - instant infrastructure in a simple, flexible, resilient,

9Copyright ©2012, Aerohive Networks, Inc.

and scalable manner, without controllers and all of their associated expenses and complexity headaches.

Figure 6 - Cooperative Control, Fully-distributed


Cooperative Control Aerohive’s Cooperative Control is a suite of protocols operating between groups of APs called hives. A hive can be as small or large as necessary and is easily customizable to fit branch office, SMB/SME, and/or large enterprise deployments. The Cooperative Control protocols manage functions such as fast/secure Layer-2 and Layer-3 roaming, coordinated RF management, Wi-Fi security, load balancing, mesh networking, and high availability. Aerohive APs are centrally-configured with a Wireless Network Management System (WNMS) called HiveManager. HiveManager is available as a hardware appliance, a virtual appliance (virtual machine), or as a Software-as-a-Service (SaaS) hosted solution called HiveManager Online. Since HiveManager is a management system used for configuration and statistics gathering, it is not essential to the network's ongoing operation. Figure 7 illustrates the fundamental building blocks of Aerohive’s Cooperative Control architecture.

Figure 7 - Fundamental Building Blocks of Cooperative Control

Economic Deficiencies of a Controller-based WLAN Architecture

When the controller-based architecture was first introduced, the expectation was that the controller-based APs would be significantly less expensive than autonomous APs. If


this turned out to be true, it would allow the total cost of the solution to be competitive in spite of the added cost of controllers. Unfortunately, this theory has not been realized in the market. Controller-based APs currently do a significant amount of processing at the AP, and they are made with the same chipsets and components as autonomous APs. As a result of this, their manufacturing costs, and hence their prices, are the same as autonomous APs. Because the controller-based architecture requires that frames be processed in two locations (the AP and/or the controller), depending on the protocol, function, or data flow, the controller-based architecture requires excessive hardware, even in optimal configurations. The cost of this additional, unnecessary hardware lands sorely in the customer’s lap, and when protocols are free, this is a great customer disservice.

Cost Variance

Controller-based deployment costs vary dramatically, depending on 1) the size of the controller, and 2) the degree to which you are able to, or want to, load the controller to its maximum AP capacity. Controller capacities rarely align with the customer’s network topology, and as a result, real-world enterprise deployments rarely hit the sweet spot in the controller cost continuum because the enterprise customer ends up buying excess controller capacity.

Redundancy

By design, the controller is the brain of a controller-based network. In many implementations, if the controller fails then all the access points either stop functioning altogether or are reduced to extremely limited capabilities. Thus the recommendation for all mission-critical implementations is for redundant controllers to be deployed. Controller redundancy, whether in a master/local, N+1, or clustered configuration, increases network availability but also significantly increases the controller component of the solution cost. This increased cost is compounded by feature licensing costs per controller.

Small Site Deployments

Small sites generally require only a few APs, sometimes only a single AP for pervasive coverage. When adding a controller to this equation, the controller can easily double the cost of hardware alone. When feature licensing is added to the equation, the controller-based architecture can become tremendously more expensive. When an organization has hundreds or perhaps thousands of remote sites, such as a large pharmacy or long-term care chain, removing controller hardware and feature licensing costs can easily reduce overall costs to a small fraction of the controller-based model.

Distributed Controllers

In demanding environments, small controllers are often deployed locally with the access points in an attempt to mitigate performance and control difficulties caused by backhauling the control and/or data plane traffic to a centrally-located controller. Distributed controllers may reduce the latency and jitter of the Wi-Fi infrastructure by reducing the number of hops that the wireless traffic must traverse to and from the controller if data flows are centralized. It also means control decisions, like firewalling and QoS, are moved closer to the network’s edge, increasing network security and the performance of applications. While this improves the operation of the controller-based


architecture in mission-critical environments, it also requires that a larger number of controllers are purchased, licensed, and deployed. This substantially drives up the cost of controller-based solutions and the complexity of the deployment.

The distributed nature of Aerohive’s Cooperative Control architecture accomplishes the same control plane tasks with zero additional costs, significantly reducing the cost of a WLAN with these requirements, and zero negative impact on data flows.

Controller Replacement and Upgrades

As an enterprise adds a new application, such as voice or video, migrates to 802.11n access points, or even adds 3-4X the number of client devices, the requirement for better performance and capacity increases. It is then likely that the capacity of an existing WLAN controller will be exceeded. In this case, additional controllers or larger replacement controllers will be needed. Assuming the enterprise requires redundancy, additional controllers would be required. An example of this would be a branch office or retail location with four APs connected to a controller capable of supporting up to six APs. If the location needs to expand to eight APs to improve RF coverage or to handle the increased number of devices on the network, doing so would require a controller replacement or an additional controller with feature licenses and the qualified manpower it would take to deploy a new Wi-Fi infrastructure system.

Over-purchasing Controller Capacity

Enterprises often purchase larger controllers than they currently need for the purpose of future network expansion. Thereafter, they need only to purchase additional APs when additional coverage or capacity is needed. While this approach is effective from a future-proofing perspective, it has a significant detrimental impact on the CAPEX of the solution. The cost impact of this scenario is especially significant when dealing with large numbers of APs or if this approach is applied to a distributed enterprise with a large number of branch or remote locations. Similarly, if the future network expansion is never performed, the additional up-front cost is wasted.

Economic Benefits of Cooperative Control Aerohive’s Cooperative Control (distributed control) approach keeps costs linear and predictable regardless of network size or deployment type, providing the capital cost advantages of autonomous APs and the technical advantages of controller-based approaches. The Cooperative Control architecture also addresses the early OPEX drawbacks of autonomous APs, especially those related to excessive administrative overhead.

Protocols Cost Less Than Controllers

Aerohive’s Cooperative Control protocols happen between clusters of APs called hives. Whether you’re talking about layer-2/3 roaming, coordinated RF management, security, load balancing, mesh networking, high-density handling, or high availability, the protocols are in there. Figure 8 illustrates the concept of protocols replacing controllers.


Figure 8 - Protocols used to distribute control - no centralized intelligence

Centralized Management with a Killer GUI and Optional SaaS Model

Centralized configuration, monitoring, and reporting are provided by Aerohive’s state-of-the-art WNMS solution called HiveManager. HiveManager isn’t your typical WNMS, and has an amazingly user-friendly GUI interface. Usability is a primary goal, and the GUI is constantly being refined to align with a network administrator’s thought processes and workflow. The HiveManager appliance, which is available in two sizes of hardware platform or a virtual appliance (virtual machine) can be located anywhere in the network and is not essential to the network’s ongoing operation.

HiveManager Online is a hosted Software-as-a-Service (SaaS) solution where HiveManger lives in the cloud, across triple-redundant data centers. HiveManager Online scales to any size, from the smallest branch office to the largest global retail chain, and it’s highly recommended for distributed enterprises. One goal of HiveManager Online is to give end-users the option of moving away from a CAPEX model and toward an OPEX model. Aerohive AP management is a modest per-AP annual fee with HiveManager Online.

Linear, Predictable, and Scalable Cost Structure

The cost structure of the Cooperative Control architecture ensures linear, predictable, and scalable costs when expanding coverage or adding capacity. Just by adding the appropriate number of Aerohive APs, organizations are able to move from a convenience-based WLAN with meeting room coverage to a mission-critical network with pervasive coverage without introducing the stair-stepped, wasteful cost model of controller-based architectures.

And it’s not just about coverage. We all understand how important capacity is. With a dual-core CPU, lots of RAM, and dual Gigabit Ethernet ports, Aerohive APs are, by a significant margin, the fastest APs in the industry. Go ahead. Throw all you’ve got at them. They won’t blink (except for the LEDs). Aerohive APs have more power and functionality than the leading controller-based architecture products, and Cooperative


Control protocols take controllers out of the equation. At the same price as controller-based APs, the superior ROI and cost model of Aerohive’s Cooperative Control architecture makes the path forward a foregone conclusion.

Inherent High Availability

The cost of redundant systems can have a huge impact on the overall WLAN solution cost, especially in large, distributed enterprises with many remote locations. The inherent, stateful high availability (HA) and mesh redundancy of the Aerohive approach rivals even the best clustered controller scenario – without the controller or licensing costs. High availability is achieved using a combination of Cooperative Control protocols and a variety of sophisticated features such as wireless mesh, fast/secure roaming, radio resource management, credential caching, and best-path forwarding. These features, and more, allow the architecture to withstand multiple Aerohive AP outages, and even a wired switch outage, without the loss of WLAN service to the user.

Green Factor

You’ve likely noticed some people’s overwhelming obsession with environmental responsibility. Large enterprise controllers are power-hungry monsters, and the amount of energy it can take to power up all of those controllers - and the air conditioning they require - is staggering. Certainly inter-AP protocols use less power than a 19” rack full of controllers, and when you factor in controller redundancy or small controllers distributed across a large number of branch offices, it’s easy to see how Aerohive’s energy-efficient APs can impact your bottom line while saving the planet. Heck, even our LEDs are green.

Cooperative Control Future-Proofs Networks The future of 802.11 is rooted in ever increasing speeds. This paper has covered the fact that the simultaneous increase in 802.11 speed coupled with the explosion of 802.11 capable devices has caused the centralized, controller-based model to break down and this problem is only getting worse. With 802.11ac on the horizon promising speeds in excess of 1Gbps, networks deployed today will need to be future-prood.

Cooperative Control’s fundamental promise is that each device can handle the clients associated with it and participate fully in the network. With a Cooperative Control architecture in your WLAN, upgrading to the next generation of WLAN does not require anything more then upgrading radios in the network. The deployment architecture remains exactly the same – no new controllers, no limitations on network capacity, no new licenses, no new security appliances. The network is architected once and it is ready for ANY wireless access method the future may hold.


Deployment Examples

Small, Single-Site Solution

The Controller-based Solution* includes one (1) 12-AP controller and ten (10) 802.11n APs with management provided by the web interface in the controller. Aerohive solution includes ten (10) 802.11n Aerohive AP 120 with full policy–based management provided by HiveManager Online.

Average, Single-Site Solution

The Controller-based Solution* includes one (1) 100-AP controller and sixty (60) 802.11n APs with management provided by the web interface in the controller. The Redundant Controller-based solution is two (2) 100-AP controllers and sixty (60) 802.11n APs with management provided by the web interface in the controller. The Aerohive solution includes sixty (60) 802.11n Aerohive AP 120 with full policy–based management provided by HiveManager Online.

$-‐

$5,000

$10,000

$15,000

Controller-‐based Aerohive

Small Single Site -‐ 10APs

Management

Controller

AP

$-‐

$50,000

$100,000

$150,000

Controller-‐based Controller-‐based with Redundancy

Aerohive

Single Site -‐ 60 APs

Management

Controller

AP


Distributed Enterprise Solution

The Controller-based Solution† includes one (1) 100-AP controller and sixty (60) 802.11n APs at HQ, and at each of the five (5) remote sites, includes one (1) 12-AP controller and ten (10) APs. Management is provided a Wireless Network Management System (WNMS). The redundant Controller-based Solution* includes two (2) 100-AP controllers and sixty (60) 802.11n APs at HQ, and at each of the five (5) remote sites, includes one (1)12-AP controller and ten (10) APs. Management is provided a Wireless Network Management System (WNMS). The Aerohive solution includes sixty (60) 802.11n Aerohive AP 120 at HQ and ten (10) 802.11n Aerohive AP 120 at each of the five (5) remote sites, with full policy–based management provided by HiveManager Online (WNMS).

Conclusion Infrastructure networks have always been fully-distributed, but due to a strong industry desire to accelerate the maturity of Wi-Fi technology within the enterprise ahead of its natural rate of progression, a centralized model was adopted. This architectural model was sufficient 10+ years ago and addressed concerns at that time.

Since the centralized control model was invented for WLANs we’ve seen the advent of the iPhone, iPad, Google Android, adoption of smartphones, Apple has become the most valuable technology company, and corporate IT must support all of this. This is a very different world from 10+ years ago.

Aerohive's fully-distributed Cooperative Control architecture has accounted for all this change and reintroduced the traditional model, which is capable of coping with tomorrow's throughput, density, resilience, cost, and scalability requirements.

Aerohive Networks’ Cooperative Control architecture provides a simple, logical, and low-cost alternative for deploying Wi-Fi infrastructures. The Aerohive approach combines a linear and predictable cost structure - regardless of deployment type or size - with the industry’s most user-friendly and scalable WNMS.

Aerohive has created a high-performance, feature-packed Wi-Fi infrastructure system that significantly reduces the overall CAPEX and OPEX costs and the complexity of deploying and scaling convenience-oriented and mission-critical enterprise networks. With Cooperative Control, the intent of Wi-Fi’s founding fathers has been realized. Protocols are free.

† List price cost comparison with the Controller-based solutions being based on Cisco’s 4404 and 4402 Controllers, 1140 series 802.11n APs, and the WCS management system.

$-‐

$50,000

$100,000

$150,000

$200,000

Controller-‐based Controller-‐based with Redundancy

Aerohive

Distributed Enterprise HQ=60APs + 5 Sites=10APs

Management

Controller

AP


About Aerohive

Aerohive Networks reduces the cost and complexity of today’s networks with cloud-enabled, distributed Wi-Fi and routing solutions for enterprises and medium sized companies including branch offices and teleworkers. Aerohive’s award-winning cooperative control Wi-Fi architecture, public or private cloud-enabled network management, routing and VPN solutions eliminate costly controllers and single points of failure. This gives its customers mission critical reliability with granular security and policy enforcement and the ability to start small and expand without limitations. Aerohive was founded in 2006 and is headquartered in Sunnyvale, Calif. The company’s investors include Kleiner Perkins Caufield & Byers, Lightspeed Venture Partners, Northern Light Venture Capital and New Enterprise Associates, Inc. (NEA).

Corporate Headquarters EMEA Headquarters Aerohive Networks, Inc. Aerohive Networks Europe LTD 330 Gibraltar Drive The Court Yard Sunnyvale, California 94089 USA 16-18 West Street Phone: 408.510.6100 Farnham Toll Free: 1.866.918.9918 Surrey, UK GU9 7DR Fax: 408.510.6199 +44 (0)1252 736590 [email protected] Fax: +44 (0)1252711901

WP1205009 www.aerohive.com xxxxxxxx

Documents

White-Paper-Next Generation Access Network Architecture...processor technology available to central controllers. It was the right tool for the problem at that time. Controllers eased