What Is Data Latency? An Expert Guide

Data latency refers to the time delay between when data is sent for use and when it produces the desired result. Latency can be caused by a number of factors, including network congestion, hardware limitations, software processing times, distance between endpoints, and how a system is configured.

Consider the response lag between hitting your car’s brake pedal and the vehicle stopping. In practice, humans tend to brake within one second of perceiving the need to do so. Imagine how destabilizing it would be if there was then a two or three second delay before the system engaged. The result would be dramatically reduced speed limits, and the safe distance between cars would need to increase greatly. Luckily, unless something is amiss, braking starts in tenths of a second, and we perceive that as an instantaneous, real-time response.

The latency in any control system—including Internet of Things (IoT) technology—will limit the precision with which it can operate. A popular IoT use case is the smart home technology that allows us to control various utilities, such as lighting, heating, security, and appliances, through a centralized smart hub or mobile application. Consider a smart thermostat. When you set it to 70 degrees, because of latency in the system, the temperature will fluctuate between about 69 and 71. But let’s say that, rather than a two or three degree fluctuation, you want only a 0.1 degree fluctuation. Because of the latency between the heat or A/C going on and the thermostat registering the change, the actual temperature would almost never be in that very narrow 0.1 degree range. So, every minute or so, the hub switches between A/C and heat. Not only would this be hard on your mechanical systems, you’d also get a rude surprise when your next energy bill arrived.

What Is Data Latency?

Data latency is the time it takes for data to travel from its source to its destination, or the time it takes for data to be processed and made available for use. This delay can be caused by various factors, including network congestion and hardware limitations, how a data collection is configured, and bottlenecks in data processing systems.

Data latency can have significant negative implications; it can also be a lever that IT can pull to balance the cost of data collection and transmission with business needs. In some industries—particularly those that require real-time or time-sensitive information, such as the financial industry—even a small delay in data transmission can lead to missed opportunities or incorrect pricing. In general, organizations will work to weigh the speed vital for decision-making and optimal performance in various applications against the costs of a faster response—and still, there’s a limit to how much latency can be driven out of a system.

While latency always degrades performance, whether a digital system becomes unusable or not depends on a range of factors.

In control systems—that is, those that manage, direct, and command other systems or devices—excessive latency causes instability and may even render the system inoperative. In real-time use cases, such as voice and video calls, latency is at best annoying and, over a certain threshold, makes the system unusable. And in data analysis, latency can slow the process to the point of rendering the finished analysis moot because decision-makers moved on without it.

Let’s look at several types of latency.

Data Latency in Real-Time Applications

Humans are very intolerant of data loss in audio and video. In audio applications, if data is delayed by more than a quarter second or so, it becomes useless and will be perceived as a click or pop or otherwise garbled audio. The same is true for video. Late data is just as bad as lost data. Algorithms can attempt to compensate, but if network latency exceeds about 300 milliseconds, real-time video will become intolerably corrupted.

In modern networks, latency is typically just tens of milliseconds, and video and audio applications run very well. Networks also recognize real-time traffic and prioritize it so that latency doesn’t render the data useless on arrival.

In near real-time applications, latency is a problem. When collecting data from sensors on a remote factory floor to monitor for malfunctions, for example, latency might slow the response time to the point of a production line shutting down before a technician can intervene. But faster information sharing requires very high performance wide area networks, which are expensive. The answer is to use so-called edge computing systems that move storage and compute closer to the machines creating the data, reducing data transport requirements, and thereby reducing latency.

Data Latency in Business Analytics

So far, the systems we’ve talked about require subsecond latency to work effectively. Business data, and the analysis of it, will always be more useful near the time it was created. But other than that very broad statement, it’s hard to say exactly how much latency is acceptable without thinking about the application at hand.

In weather forecasting, data on last week’s atmospheric conditions might have some use in identifying weather patterns for this week, but newer data from the last few days and hours will be much more useful. Still, there’s a limit to how much timeliness matters. Do weather forecasters need data updates each second or each tenth of a second? At some point, there just isn’t enough variation in data collected over very short intervals to be worth the expense.

The same sort of thinking is important when calculating how much latency is acceptable in the design of business analysis systems. Does a large retailer want to know how many blue shirts sell on a second-by-second basis? Probably not. Is hourly sales and inventory data sufficient? Considering how difficult it would be to make a meaningful change in inventory in under an hour, yes, that probably is sufficient. Attaining better performance would cost money that might be better spent elsewhere.

On the other hand, reports on sales data that’s accurate to within a few hours or even a couple of days could be quite valuable. Executives might want to know what items people who purchase blue shirts also buy, even if it’s not in the same shopping instance. They could create sales bundles—think “Buy the look” offers—based on this data. Likewise, online shoppers appreciate “People who bought this, also bought that” recommendations. So the latency in collecting “People also bought” data might be much more critical than simply how many shirts sold. The difference between having store managers’ reports land in their inboxes every day or so and letting them pull up the data in seconds can be a real business game changer.

So, while latency affects how data is collected, how it’s processed, and how it’s made available for future analysis, deciding which area to improve should start with an understanding of the business challenge. Data collection needs to be as fast as the business needs, but spending money to make it faster could be an unwise investment. The key is to focus on areas where new technology investments will have the biggest impact.

Meanwhile, moving from a system that requires an ETL process before analysis can begin to one that allows data to be stored in an advanced database environment could let business leaders understand more trends more quickly—possibly even in real time. Adding self-service capabilities so business leaders can generate common analyses will also reduce another source of latency.

This ties into the concept of business latency, the time between when an unexpected event affecting future performance happens to the moment in which your organization acts on this information.

It is crucial to manage latency because it can greatly impact user experience and satisfaction. High latency can lead to slow loading times, delays in processing requests, and overall poor performance. Monitoring and optimizing latency levels are worth the effort. By reducing latency, businesses can improve customer and employee satisfaction.

Data Latency FAQs

What is acceptable data latency?

Acceptable data latency varies based on the organization or system utilizing the data. In general, acceptable data latency refers to the maximum amount of time it can take for data to be transmitted or processed from its source to its destination to deliver acceptable performance. In some cases, such as in financial systems or mission-critical applications, real-time data processing with minimal latency is crucial. In other cases, such as in data analytics or reporting where data updates can be batched, a slight delay in data transmission or processing may be acceptable. Ultimately, what constitutes “acceptable” data latency is determined by the specific priorities and use cases of the organization or system.

In servers, latency between memory, CPU, and network adaptors will be measured in microseconds. In large-scale storage systems, it’s milliseconds. When customers are making purchases, committing transactions to storage might take a fraction of a second or so. And when humans are involved, latency can be much longer. Determining acceptable latency will almost always depend on the application at hand.

How do you handle data latency?

Strategies for handling data latency depend on the origin of latency. If an organization only sees a handful of complaints regarding latency, the issue may stem from the user’s side. Possible causes include outdated devices or a slow internet connection. For widespread issues, this is a sign that the cause rests on the organization side. A suite of best practices to help fix data latency can include implementing caching tools, monitoring tools, better data compression strategies, and considering a better cloud infrastructure provider.

Does data rate affect latency?

Data rate refers to the speed at which data is sent across a network, usually expressed in bytes per second. Data latency is related to this, as it is the time gap between request and response. Higher data rates can help to reduce data latency, as higher data rates ensure better bandwidth and overall performance. However, data latency is not necessarily directly dependent on data rate, as other variables may apply. They are related enough, though, that data rate issues may be indicative of latency problems and vice versa.

How can latency variation be reduced?

For users, latency variation can stem from something as simple as unreliable internet connection or low memory/storage on a device; addressing those issues usually clears up the variation. On the provider side, latency variation can be a symptom of sudden compute requests eating up processing power or bandwidth. End-to-end application monitoring as well as general network monitoring should provide insight as to why sudden latency spikes occur. Once the root cause is isolated, IT teams can implement optimization strategies.