Bundling inexpensive Internet links to build reliable, enterprise-class WANs

WAN virtualization tools can lead to 40% to 90% savings

Data center centralization and consolidation. Cloud computing. Latency-sensitive (real-time and interactive) applications such as VoIP, videoconferencing and virtual desktop infrastructures (VDI). Business continuity and disaster recovery. These enterprise trends are among those driving the need for a WAN access layer that is scalable, reliable and cost-effective.

While MPLS services from service providers such as AT&T, Verizon and BT meet the "four nines" (99.99%) reliability requirements enterprises expect, they are extremely expensive. At typical U.S. domestic pricing of ~$400 per megabit per month for copper T1-based connections for the typical remote office, companies have simply been unable to scale bandwidth at anywhere near the rate at which demand is growing.

WAN optimization tools offer a way to delay bandwidth upgrades, and can perform wonders for application acceleration, but you can only put off the need for additional bandwidth for so long, especially as multimedia file sizes continue to increase and bandwidth-hogging applications like video become ever more popular.

Public Internet bandwidth is significantly less costly -- typically $10-$15 per megabit per month for ADSL, sub $4 per megabit per month for cable, and as low as $5 per megabit per month for high-speed bandwidth at carrier-neutral colocation facilities -- but these services are only about 99% reliability.

This is simply not reliable enough for most corporate WAN requirements. The popular perception of the Internet is indeed accurate: Unaided, the Internet works pretty well most of the time. But "pretty well" is not good enough for most enterprises, and "most of the time" is not good enough for almost any. Note that when we refer to reliability here, we mean the union of simple availability with whether packets are getting through to their destination without being lost or excessively delayed.

However, if one remembers their basic systems' reliability course lesson, while components which are each 99% reliable run in series deliver only 98% system reliability, the same components run in parallel, in a properly designed system, will deliver 99.99% reliability.

A new technology called WAN Virtualization combines diverse IP WAN circuits at each location, and makes it possible to get 30 to 100 times the bandwidth per dollar, 40% to 90% WAN cost savings, and more reliability and application performance predictability than the best, single-vendor MPLS network.

Reliability chart

Where server virtualization leverages the efficient pooling of computing and storage resources, WAN Virtualization does a similar pooling of wide area network resources. An appliance-based WAN Virtualization solution typically wraps a layer of hardware and intelligent software around multiple WAN connections -- existing private links such as MPLS, as well as any kind of Internet WAN link, be it DSL, cable, fiber, Metro Ethernet, etc. -- to augment or replace private WAN connections, delivering not only more bandwidth and lower monthly cost, but greater reliability and application performance predictability.

How WAN Virtualization works

Well designed WAN Virtualization solutions are of necessity two-ended, and are usually appliance-based. They require two (or more) ISP pipes to each location, or as little as an existing MPLS connection plus the local Internet access/VPN backup link. Of course, aggregating several links per location, many of which are from the same service provider, is also beneficial to allow additional bandwidth.

The appliances then support per-flow classification and QoS across these aggregated connections, which typically support IPSec or 128-bit AES encryption. To deliver maximum application performance and performance predictability, the best WAN Virtualization solutions will make per-packet, rather than per-flow, forwarding decisions, enabling, among other benefits, the use of all available bandwidth even for a single TCP flow.

A WAN Virtualization solution should conduct continuous measurement of the state of each network path in each direction: loss, latency, jitter and bandwidth utilization, preferably many times per second. Unidirectional measurement, rather than simply measuring these statistics on a round-trip basis, is important to ensure correct handling of network congestion events.

Since packet loss is the biggest killer of IP application performance, the WAN Virtualization tools need to do their best to avoid loss and mitigate its effects when it occurs through techniques including buffering, retransmission, re-ordering, selective replication, etc.

The key to a good WAN Virtualization solution is the ability to react sub-second to severe problems with any network connection -- not just link failures, but also congestion-related network problems. If the appliances can do dynamic, real-time traffic engineering, enterprises can use all of the bandwidth they're paying for most of the time as well as most of that bandwidth all of the time, with as good -- and usually better -- performance predictability as the best MPLS network alone can deliver.

Advanced WAN Virtualization solutions, in addition to adapting around congestion which occurs in the middle of any network, will incorporate techniques to avoid last-mile congestion, increasing network utilization and efficiency while delivering better applications performance. They will ensure that not only real-time traffic like VoIP, but also interactive traffic like VDI or Web-based applications, is continuously put on the best performing network paths, rather than simply being placed on the connection that is "usually" better at session initiation, and moved only in the event of a complete link failure.

WAN Virtualization can be added seamlessly to existing networks and can be deployed one site at a time or on an application, user or server basis until the buyer is confident in the reliability.

Similar to existing WAN Optimization solutions, WAN Virtualization deployments are typically done as a network overlay, where the appliances can be deployed either in-line or out-of-line, and support both fail-to-wire capability and high availability redundancy options for both ease of deployment and maximum reliability.

In fact, WAN Virtualization is complementary to WAN Optimization technology, which is already widely deployed for the application acceleration benefits. The bandwidth aggregation and loss mitigation capabilities of WAN Virtualization deliver added benefits to WAN Optimization, providing better throughput for first-time data transfers, more predictable performance for all applications when any part of the WAN becomes congested, and reliable performance for real-time applications -- such as VoIP and videoconferencing -- for which WAN Optimization technology cannot do very much.

To meet the reliability, scalability and affordability needs of network applications as data center consolidation and cloud computing become more prevalent, an alternative to paying high prices for private MPLS connections is needed. WAN Virtualization technology offers enterprise WAN managers the ability to cost effectively leverage Internet bandwidth to augment, and optionally replace, expensive MPLS connections, without sacrificing reliability or performance predictability.

With WAN Virtualization, Talari Networks does for enterprise WANs what RAID did for storage -- delivering a network with 30 to 100 times the bandwidth per dollar, reduced ongoing WAN costs by 40% to 90%, and greater reliability than existing corporate WANs -- bringing Moore's Law and Internet economics to enterprise WAN buyers for the first time in more than 15 years.

Learn more about this topic

The ABCs of WAN optimization savings

Users find the secrets of WAN optimization

Will videoconferencing replace the telephone?

Join the Network World communities on Facebook and LinkedIn to comment on topics that are top of mind.

Copyright © 2011 IDG Communications, Inc.