Oct 12, 2012
When discussing cloud computing, the term “East-West latency” often comes up as users on one coast of the US have found themselves using data centres on the opposite coast – and find that the user experience was not quite as they would wish.
The direct distance between San Diego, California on the western seaboard to Bangor, Maine on the eastern is around 2,700 miles. At a theoretical level, the latency of a speed of light connection should therefore be around 15 milliseconds, or 30 milliseconds for a round trip. However, due to the fact that the connection is not straight line and other issues with packet routing, jitter and so on, the round-trip latency tends to be closer to between 100 and 200 milliseconds.
With voice and video becoming part of the data network, even low latency issues are picked up by the human brain. Such small lags in response can make the difference between a successful implementation of an application and a failed one.
However, technology is rapidly transcending the physical borders of given countries. Many larger organisations now need data centres that support a global user base. Therefore, an organisation may not only have data centres in New York and San Francisco, but also in Tokyo, Sydney, Bangalore, London and Johannesburg. Round trips for data may be a matter of 20,000 miles or more – with latencies being measurable in the order of half a second or more. For end users, application response times will be intolerable, forcing them to find other ways to carry out their work.
This puts IT departments – and the co-location and cloud data centre providers they partner with – in a difficult position: just how can a global organisation create and run a seamless IT platform when the speed of light seems to be against them?
First, the application architecture needs to be implemented correctly. Wherever possible, the business logic and the data should be kept in close proximity. If large amounts of data must traverse paths between multiple data centres dotted around the planet, response time will always be degraded. Keeping as much of the traffic as possible within a single data centre will mean the capability to use the higher bandwidth of the local area network (LAN) and attain network speeds with close to zero internal latency. If done effectively, only the visualization of the results needs to be sent to the user, resulting in far less traffic, and then wide area network (WAN) acceleration can be applied to further enhance the experience.
Next, the back end of the enterprise systems still needs managing. A single data centre approach does not provide for high availability. To achieve this, data should be mirrored across data centres to provide needed redundancy. By using data deduplication and WAN optimisation, this can be done in near real time, and the combined use of store and forward technologies can make the recovery point objective (RPO) measurable in terms of seconds or minutes.
And what about voice and video? Prioritisation of real-time streams can be managed through priority and quality of service approaches (QoS), such as using packet marking through 802.1p/q or multi-protocol labelling switching (MPLS). However, this may not be enough, and additional WAN optimisation services, such as packet shaping, may be needed to ensure that voice and video meet mean opinion score (MOS) metrics.
As globalisation increases, latency will grow as an issue. However, although the laws of physics cannot be broken – they can be bent, with the right WAN optimisation tools.
Image source: flickr (woodleywonderworks)