Run your network like your business with intelligent TCP flow control

Today's WAN infrastructure is unable to deliver the performance characteristics demanded by businesses. Learn how to turn that around with intelligent TCP flow control.

Paul Sutter

 In today's competitive economy, businesses measure success in terms of their ability to quickly and effectively react to changes in market conditions and customer trends. They require predictable performance from their business processes, and demand risk-free response time guarantees from their critical application workflows. This behavior is expected to hold true at any geographical distance, across all network conditions.

Unfortunately, today's WAN infrastructure is unable to deliver the performance characteristics demanded by businesses. This performance instability stems from the inadequate flow control algorithms in use by Transmission Control Protocol (TCP), which is used by a large number of business-critical applications traversing WANs today. Vendors in the WAN optimization space have traditionally used techniques such as compression, caching, and QoS to cover for TCP's limitations. These techniques solve a myriad of problems faced by application workflows, but merely serve as a band-aid for TCP's inherent limitations. Intelligent TCP flow control (ITFC) can contribute to WAN optimization solution, and is responsible for adding the predictability factor to TCP-based workflows, guaranteeing optimally streamlined business processes that make full use of the WAN infrastructure at any distance, over any network conditions.

TCP was designed to simplify application interaction over LANs. Each TCP session in the network is unaware of any other TCP session in the system, and relies upon various permutations of packet loss and timeouts to detect and adjust for network anomalies. Unfortunately, the detection and correction mechanisms used by each TCP session are collectively unable to provide the throughput and delivery time guarantees needed for business operations. In addition, as the distance between session endpoints increases, the algorithms become increasingly unfair

TCP sessions are analogous to high-speed bumper cars that are following a known course in zero-visibility conditions. When the bumper cars bump into each other, they slow down to half their respective speeds and then slowly build up speed until they bump into someone else. Because of this, they never realize their true potential. ITFC puts an end to the bump-and-go status quo by planning for and dictating the speed or send rate at which each TCP session should operate.

ITFC operates at either end of a WAN link, and is typically deployed at network transition points, giving it a higher level of network visibility. Mature ITFC implementations are able to differentiate between bump-and-go behavior and true network congestion anomalies and are able to adjust the send rate for each TCP session in the system accordingly. Assuming a predetermined send rate for the entire WAN link, ITFC can enforce a fair (or configured) send rate per session, regardless of the network distance between TCP end points. This makes it easy for managers to plan data transfers across WAN links. Advanced ITFC implementations are able to carry out all TCP optimizations transparently, requiring zero network reconfiguration and zero changes to the application infrastructure. TCP sessions are therefore sheltered from all network anomalies, and can deliver unmatched and predictable performance anytime, anywhere.

The effective throughput attained by a TCP session is dependent on three unique variables:


  1. Network latency, the distance between two TCP end-points
  2. Packet loss, the average number of packets lost due to network conditions
  3. Maximum segment size (MSS), the amount of data that may be exchanged between the endpoints per transfer cycle.

WAN optimization techniques such as compression, caching and QoS improve TCP performance by effectively increasing the MSS per TCP session, thereby applying a multiplicative effect on the amount of data being transferred. However, they do not solve any of the issues that arise due to latency and loss. A slight change in either of these can lead to vastly different throughput results. Also note that TCP's native flow control algorithm throttles the MSS when network anomalies are detected, further minimizing the performance improvement provided by compression, caching, QoS, etc. ITFC effectively shelters all TCP end points from the travails of dealing with latency and loss, leading to predictable throughput for all TCP sessions in the system. ITFC solutions can also be configured to use extremely large MSS values, further improving TCP performance.

ITFC can also be used in conjunction with other WAN optimization techniques. True WAN optimization uses several different techniques, each leveraging the other to deliver optimum performance. ITFC works to nullify the impact of network distances and adverse conditions, allowing techniques such as compression, caching and QoS to apply a multiplicative effect on throughput rates. Based on the application type, network conditions or configured policies, WAN optimization can also select the techniques best suited for the workflow in question, making previously complex (and sometimes impossible) application optimizations a piece of cake.

Financial planning and predictability are sure signs of success for any business. This predictability has been missing from business-critical application workflows but is now possible with WAN optimization. Business managers can expect the same levels of pre-planning and careful execution with their network infrastructure deployments that they do with their day-to-day business strategies. They can now invest in running their networks like their businesses, expecting and achieving hard returns for each dollar spent to support application workflows.

About the author:
Paul Sutter is Founder and Chief Technology Officer at Orbital Data, where he is responsible for the technical vision and direction for the company's product line. Prior to Orbital, Paul was vice president of engineering at AltaVista where he was instrumental in the development of the company's portal and search engine businesses. Paul came to AltaVista through its acquisition of Transium, an ASP hosting company, where Paul was the founder and chief executive officer, growing the business to include customers such as Yahoo!, and over three years. Previously, Paul founded Voila! Software and started his career at Apple Computer in the company's networking and communications group.

This was last published in February 2005

Dig Deeper on Network Infrastructure

Start the conversation

Send me notifications when other members comment.

Please create a username to comment.