How to Improve Effective Throughput

It may surprise you to learn that more bandwidth does not necessarily mean higher effective WAN throughput. If you transfer a lot of data over a high-capacity WAN, we bet you are not getting the effective throughput you think you are--and upgrading bandwidth won't help because technical and physical constraints stack the cards against you. We describe why this is and what you can do to about it in a free report you can download here--and Peter Sevcik will discuss it in an upcoming Webinar. Here's a quick synopsis.

What Is Effective Throughput?

To figure out how effectively you use WAN capacity for large data transfers, you need to know how much available capacity you actually put to work. To understand how much of that capacity you actually use, you must first determine your effective throughput--the number of bits per second successfully delivered from source to destination for an individual data flow. The higher your effective throughput, the more efficiently you use your WAN capacity.

Once you know your effective throughput, you need to analyze it relative to the total available bandwidth by calculating the effective throughput ratio (ETR). This entails dividing effective throughput by available bandwidth between source and destination (typically the WAN access circuit). The effective throughput ratio tells you how much of the available bandwidth your single flow uses.

The effective throughput ratio for a single flow is highest near the data source and decreases with distance. This is orthogonal to bandwidth utilization, which is a measure of capacity consumed by many flows. Unlike the effective throughput ratio, bandwidth utilization typically starts low and increases with the number of simultaneous users on the path.

What Makes Effective Throughput Poor?

Three factors conspire to erode effective WAN throughput: distance, TCP window size, and packet loss. These factors make effective throughput lower than the bandwidth of the slowest link along the path--usually an access link. The reason for this hinges on the Automatic Repeat Request (ARQ) mechanism within TCP.

ARQ uses a sliding window to enable the sender to transmit multiple packets before waiting for an acknowledgement from the receiver. A faster circuit puts a window's worth of data in flight faster and then must wait longer before sending more, so the percentage of idle time actually increases compared to a slower circuit. The larger the circuit, the more dramatically the ACK wait time lowers effective throughput because no data can be transmitted while both the data and the ACK are in flight.

What Can You Do about It?

Help is available in the form of WAN optimization solutions. The best of these solutions include TCP optimization to maximize TCP window size, forward error correction to address packet loss, and packet order correction to fix out-of-order packets, which also lowers packet loss.

TCP optimization increases TCP window size, which puts more data in flight on long latency paths. TCP optimization can also include a variety of actions such as sending pre-emptive data receipt acknowledgements that maintain high throughput to speed data from the source, and ramping up the TCP transmission rate more quickly by bypassing TCP's ‘slow start' function. TCP optimization also uses a selective acknowledgement (SACK) feature that retransmits only bytes lost rather than returning to the last continuously received data.

Forward error correction fixes errors in real time, avoiding the need to retransmit data when packets are lost. Although forward error correction adds overhead, the benefits make the tradeoff acceptable for underutilized high-capacity WAN links. Some WAN optimization solutions minimize overhead by dynamically matching forward error correction levels to loss levels.

MPLS and IP VPN environments routinely suffer from out-of-order packet delivery. TCP identifies more than three packets received out of order as packet loss and calls for packet retransmissions and smaller TCP window size. This response can be particularly vexing when trying to keep many bytes in flight using a large window size. Packet order correction properly sequences out-of-order packets on the fly, thus avoiding retransmissions.

The Upshot

The report shows extensive analysis that models the before and after results achievable using TCP optimization.  When you combine all the WAN optimization technologies described above you can improve effective throughput ratios by 5x to 10x on average, with peaks as high as 50x. This makes WAN optimization solutions a very good way (in fact the best way we know of) to improve your effective throughput.

         Download the Report.

Join the Network World communities on Facebook and LinkedIn to comment on topics that are top of mind.
Must read: 10 new UI features coming to Windows 10