Another factor in network delay that is often overlooked is the need to correctly size bandwidth against the traffic profile of the circuit. In a trading environment, the actual bandwidth required is highly dependent on the application and can vary dramatically from minute to minute throughout the trading day. Ignoring this can result in queuing delay, a situation in which a data bottleneck develops because the circuit does not have sufficient bandwidth to support bursts in data.
For example, the traffic profile of a market data feed from an exchange exhibits relatively low data rates throughout most of the day, punctuated by spikes at market open, end-of-day, and immediately following the release of market-moving news stories. These data spikes can easily reach peak traffic levels that are three or five times the average if a significant market event occurs.
If sufficient bandwidth is unavailable for the peak traffic caused by these bursts of traffic, then data packets backup until the queue is cleared. Should the data queue become too great, packets are dropped and re-transmitted. In either case, the result is significant delays in the transmission of data. Invariably, these market data spikes – and the resultant delays – occur at times of market volatility, when timely data is most critical.
If the number of delayed or dropped packets is low, the slight increase in latency may not be readily apparent to market data displays or application monitoring tools; however it could considerably reduce the effectiveness of a program trading application.