Welcome!

Related Topics: Java IoT, Microservices Expo, Adobe Flex, IoT User Interface, Apache

Java IoT: Article

Why Averages Are Inadequate, and Percentiles Are Great

Averages are ineffective because they are too simplistic and one-dimensional

Anyone who ever monitored or analyzed an application uses or has used averages. They are simple to understand and calculate. We tend to ignore just how wrong the picture is that averages paint of the world. To emphasis the point let me give you a real-world example outside of the performance space that I read recently in a newspaper.

The article was explaining that the average salary in a certain region in Europe was 1900 Euro's (to be clear this would be quite good in that region!). However when looking closer they found out that the majority, namely 9 out of 10 people, only earned around 1000 Euros and one would earn 10.000 (I over simplified this of course, but you get the idea). If you do the math you will see that the average of this is indeed 1900, but we can all agree that this does not represent the "average" salary as we would use the word in day to day live. So now let's apply this thinking to application performance.

The Average Response Time
The average response time is by far the most commonly used metric in application performance management. We assume that this represents a "normal" transaction, however this would only be true if the response time is always the same (all transaction run at equal speed) or the response time distribution is roughly bell curved.

A Bell curve represents the "normal" distribution of response times in which the average and the median are the same. It rarely ever occurs in real applications

In a Bell Curve the average (mean) and median are the same. In other words observed performance would represent the majority (half or more than half) of the transactions.

In reality most applications have few very heavy outliers; a statistician would say that the curve has a long tail. A long tail does not imply many slow transactions, but few that are magnitudes slower than the norm.

This is a typical Response Time Distribution with few but heavy outliers - it has a long tail. The average here is dragged to the right by the long tail.

We recognize that the average no longer represents the bulk of the transactions but can be a lot higher than the median.

You can now argue that this is not a problem as long as the average doesn't look better than the median. I would disagree, but let's look at another real-world scenario experienced by many of our customers:

This is another typical Response Time Distribution. Here we have quite a few very fast transactions that drag the average to the left of the actual median

In this case a considerable percentage of transactions are very, very fast (10-20 percent), while the bulk of transactions are several times slower. The median would still tell us the true story, but the average all of a sudden looks a lot faster than most of our transactions actually are. This is very typical in search engines or when caches are involved - some transactions are very fast, but the bulk are normal. Another reason for this scenario are failed transactions, more specifically transactions that failed fast. Many real-world applications have a failure rate of 1-10 percent (due to user errors or validation errors). These failed transactions are often magnitudes faster than the real ones and consequently distorted an average.

Of course performance analysts are not stupid and regularly try to compensate with higher frequency charts (compensating by looking at smaller aggregates visually) and by taking in minimum and maximum observed response times. However we can often only do this if we know the application very well, those unfamiliar with the application might easily misinterpret the charts. Because of the depth and type of knowledge required for this, it's difficult to communicate your analysis to other people - think how many arguments between IT teams have been caused by this. And that's before we even begin to think about communicating with business stakeholders!

A better metric by far are percentiles, because they allow us to understand the distribution. But before we look at percentiles, let's take a look a key feature in every production monitoring solution: Automatic Baselining and Alerting.

Automatic Baselining and Alerting
In real-world environments, performance gets attention when it is poor and has a negative impact on the business and users. But how can we identify performance issues quickly to prevent negative effects? We cannot alert on every slow transaction, since there are always some. In addition, most operations teams have to maintain a large number of applications and are not familiar with all of them, so manually setting thresholds can be inaccurate, quite painful and time-consuming.

The industry has come up with a solution called Automatic Baselining. Baselining calculates out the "normal" performance and only alerts us when an application slows down or produces more errors than usual. Most approaches rely on averages and standard deviations.

Without going into statistical details, this approach again assumes that the response times are distributed over a bell curve:

The Standard Deviation represents 33% of all transactions with the mean as the middle. 2xStandard Deviation represents 66% and thus the majority, everything outside could be considered an outlier. However most real world scenarios are not bell curved...

Typically, transactions that are outside two times standard deviation are treated as slow and captured for analysis. An alert is raised if the average moves significantly. In a bell curve this would account for the slowest 16.5 percent (and you can of course adjust that); however; if the response time distribution does not represent a bell curve, it becomes inaccurate. We either end up with a lot of false positives (transactions that are a lot slower than the average but when looking at the curve lie within the norm) or we miss a lot of problems (false negatives). In addition if the curve is not a bell curve, then the average can differ a lot from the median; applying a standard deviation to such an average can lead to quite a different result than you would expect. To work around this problem these algorithms have many tunable variables and a lot of "hacks" for specific use cases.

Why I Love Percentiles
A percentile tells me which part of the curve I am looking at and how many transactions are represented by that metric. To visualize this look at the following chart:

This chart shows the 50th and 90th percentile along with the average of the same transaction. It shows that the average is influenced far mor heavily by the 90th, thus by outliers and not by the bulk of the transactions

The green line represents the average. As you can see it is very volatile. The other two lines represent the 50th and 90th percentile. As we can see the 50th percentile (or median) is rather stable but has a couple of jumps. These jumps represent real performance degradation for the majority (50%) of the transactions. The 90th percentile (this is the start of the "tail") is a lot more volatile, which means that the outliers slowness depends on data or user behavior. What's important here is that the average is heavily influenced (dragged) by the 90th percentile, the tail, rather than the bulk of the transactions.

If the 50th percentile (median) of a response time is 500ms that means that 50% of my transactions are either as fast or faster than 500ms. If the 90th percentile of the same transaction is at 1000ms it means that 90% are as fast or faster and only 10% are slower. The average in this case could either be lower than 500ms (on a heavy front curve), a lot higher (long tail) or somewhere in between. A percentile gives me a much better sense of my real world performance, because it shows me a slice of my response time curve.

For exactly that reason percentiles are perfect for automatic baselining. If the 50th percentile moves from 500ms to 600ms I know that 50% of my transactions suffered a 20% performance degradation. You need to react to that.

In many cases we see that the 75th or 90th percentile does not change at all in such a scenario. This means the slow transactions didn't get any slower, only the normal ones did. Depending on how long your tail is the average might not have moved at all in such a scenario.!

In other cases we see the 98th percentile degrading from 1s to 1.5 seconds while the 95th is stable at 900ms. This means that your application as a whole is stable, but a few outliers got worse, nothing to worry about immediately. Percentile-based alerts do not suffer from false positives, are a lot less volatile and don't miss any important performance degradations! Consequently a baselining approach that uses percentiles does not require a lot of tuning variables to work effectively.

The screenshot below shows the Median (50th Percentile) for a particular transaction jumping from about 50ms to about 500ms and triggering an alert as it is significantly above the calculated baseline (green line). The chart labeled "Slow Response Time" on the other hand shows the 90th percentile for the same transaction. These "outliers" also show an increase in response time but not significant enough to trigger an alert.

Here we see an automatic baselining dashboard with a violation at the 50th percentile. The violation is quite clear, at the same time the 90th percentile (right upper chart) does not violate. Because the outliers are so much slower than the bulk of the transaction an average would have been influenced by them and would not have have reacted quite as dramatically as the 50th percentile. We might have missed this clear violation!

How Can We Use Percentiles for Tuning?
Percentiles are also great for tuning, and giving your optimizations a particular goal. Let's say that something within my application is too slow in general and I need to make it faster. In this case I want to focus on bringing down the 90th percentile. This would ensure sure that the overall response time of the application goes down. In other cases I have unacceptably long outliers I want to focus on bringing down response time for transactions beyond the 98th or 99th percentile (only outliers). We see a lot of applications that have perfectly acceptable performance for the 90th percentile, with the 98th percentile being magnitudes worse.

In throughput oriented applications on the other hand I would want to make the majority of my transactions very fast, while accepting that an optimization makes a few outliers slower. I might therefore make sure that the 75th percentile goes down while trying to keep the 90th percentile stable or not getting a lot worse.

I could not make the same kind of observations with averages, minimum and maximum, but with percentiles they are very easy indeed.

Conclusion
Averages are ineffective because they are too simplistic and one-dimensional. Percentiles are a really great and easy way of understanding the real performance characteristics of your application. They also provide a great basis for automatic baselining, behavioral learning and optimizing your application with a proper focus. In short, percentiles are great!

More Stories By Michael Kopp

Michael Kopp has over 12 years of experience as an architect and developer in the Enterprise Java space. Before coming to CompuwareAPM dynaTrace he was the Chief Architect at GoldenSource, a major player in the EDM space. In 2009 he joined dynaTrace as a technology strategist in the center of excellence. He specializes application performance management in large scale production environments with special focus on virtualized and cloud environments. His current focus is how to effectively leverage BigData Solutions and how these technologies impact and change the application landscape.

Comments (1) View Comments

Share your thoughts on this story.

Add your comment
You must be signed in to add a comment. Sign-in | Register

In accordance with our Comment Policy, we encourage comments that are on topic, relevant and to-the-point. We will remove comments that include profanity, personal attacks, racial slurs, threats of violence, or other inappropriate material that violates our Terms and Conditions, and will block users who make repeated violations. We ask all readers to expect diversity of opinion and to treat one another with dignity and respect.


Most Recent Comments
rtalexander 11/21/12 12:58:00 AM EST

Hey, could you post a reference or two that covers the theory and/or practicalities of the approach you describe?

Thanks!

Latest Stories
"We host and fully manage cloud data services, whether we store, the data, move the data, or run analytics on the data," stated Kamal Shannak, Senior Development Manager, Cloud Data Services, IBM, in this SYS-CON.tv interview at 18th Cloud Expo, held June 7-9, 2016, at the Javits Center in New York City, NY.
Machine Learning helps make complex systems more efficient. By applying advanced Machine Learning techniques such as Cognitive Fingerprinting, wind project operators can utilize these tools to learn from collected data, detect regular patterns, and optimize their own operations. In his session at 18th Cloud Expo, Stuart Gillen, Director of Business Development at SparkCognition, discussed how research has demonstrated the value of Machine Learning in delivering next generation analytics to imp...
DevOps at Cloud Expo – being held November 1-3, 2016, at the Santa Clara Convention Center in Santa Clara, CA – announces that its Call for Papers is open. Born out of proven success in agile development, cloud computing, and process automation, DevOps is a macro trend you cannot afford to miss. From showcase success stories from early adopters and web-scale businesses, DevOps is expanding to organizations of all sizes, including the world's largest enterprises – and delivering real results. Am...
With the proliferation of both SQL and NoSQL databases, organizations can now target specific fit-for-purpose database tools for their different application needs regarding scalability, ease of use, ACID support, etc. Platform as a Service offerings make this even easier now, enabling developers to roll out their own database infrastructure in minutes with minimal management overhead. However, this same amount of flexibility also comes with the challenges of picking the right tool, on the right ...
The 19th International Cloud Expo has announced that its Call for Papers is open. Cloud Expo, to be held November 1-3, 2016, at the Santa Clara Convention Center in Santa Clara, CA, brings together Cloud Computing, Big Data, Internet of Things, DevOps, Digital Transformation, Microservices and WebRTC to one location. With cloud computing driving a higher percentage of enterprise IT budgets every year, it becomes increasingly important to plant your flag in this fast-expanding business opportuni...
With over 720 million Internet users and 40–50% CAGR, the Chinese Cloud Computing market has been booming. When talking about cloud computing, what are the Chinese users of cloud thinking about? What is the most powerful force that can push them to make the buying decision? How to tap into them? In his session at 18th Cloud Expo, Yu Hao, CEO and co-founder of SpeedyCloud, answered these questions and discussed the results of SpeedyCloud’s survey.
The Internet of Things will challenge the status quo of how IT and development organizations operate. Or will it? Certainly the fog layer of IoT requires special insights about data ontology, security and transactional integrity. But the developmental challenges are the same: People, Process and Platform. In his session at @ThingsExpo, Craig Sproule, CEO of Metavine, demonstrated how to move beyond today's coding paradigm and shared the must-have mindsets for removing complexity from the develo...
SYS-CON Events announced today that MangoApps will exhibit at the 19th International Cloud Expo, which will take place on November 1–3, 2016, at the Santa Clara Convention Center in Santa Clara, CA. MangoApps provides modern company intranets and team collaboration software, allowing workers to stay connected and productive from anywhere in the world and from any device.
Large scale deployments present unique planning challenges, system commissioning hurdles between IT and OT and demand careful system hand-off orchestration. In his session at @ThingsExpo, Jeff Smith, Senior Director and a founding member of Incenergy, will discuss some of the key tactics to ensure delivery success based on his experience of the last two years deploying Industrial IoT systems across four continents.
Redis is not only the fastest database, but it is the most popular among the new wave of databases running in containers. Redis speeds up just about every data interaction between your users or operational systems. In his session at 19th Cloud Expo, Dave Nielsen, Developer Advocate, Redis Labs, will share the functions and data structures used to solve everyday use cases that are driving Redis' popularity.
Predictive analytics tools monitor, report, and troubleshoot in order to make proactive decisions about the health, performance, and utilization of storage. Most enterprises combine cloud and on-premise storage, resulting in blended environments of physical, virtual, cloud, and other platforms, which justifies more sophisticated storage analytics. In his session at 18th Cloud Expo, Peter McCallum, Vice President of Datacenter Solutions at FalconStor, discussed using predictive analytics to mon...
Basho Technologies has announced the latest release of Basho Riak TS, version 1.3. Riak TS is an enterprise-grade NoSQL database optimized for Internet of Things (IoT). The open source version enables developers to download the software for free and use it in production as well as make contributions to the code and develop applications around Riak TS. Enhancements to Riak TS make it quick, easy and cost-effective to spin up an instance to test new ideas and build IoT applications. In addition to...
Identity is in everything and customers are looking to their providers to ensure the security of their identities, transactions and data. With the increased reliance on cloud-based services, service providers must build security and trust into their offerings, adding value to customers and improving the user experience. Making identity, security and privacy easy for customers provides a unique advantage over the competition.
In his session at @DevOpsSummit at 19th Cloud Expo, Yoseph Reuveni, Director of Software Engineering at Jet.com, will discuss Jet.com's journey into containerizing Microsoft-based technologies like C# and F# into Docker. He will talk about lessons learned and challenges faced, the Mono framework tryout and how they deployed everything into Azure cloud. Yoseph Reuveni is a technology leader with unique experience developing and running high throughput (over 1M tps) distributed systems with extre...
"We've discovered that after shows 80% if leads that people get, 80% of the conversations end up on the show floor, meaning people forget about it, people forget who they talk to, people forget that there are actual business opportunities to be had here so we try to help out and keep the conversations going," explained Jeff Mesnik, Founder and President of ContentMX, in this SYS-CON.tv interview at 18th Cloud Expo, held June 7-9, 2016, at the Javits Center in New York City, NY.