APM at the Scale of Business: Delivering App Insights with No Blind Spots


Modern applications are increasingly complex, and that's why enterprises are turning to application performance monitoring (APM) platforms to help troubleshoot and optimize their infrastructure. But the reality is that many APM tools are overwhelmed by all that complexity, and unable to provide you with the insights you need.

SteelCentral is the only APM solution that doesn't blink when faced with big data, providing all the information needed to monitor even the most hair-raisingly complex enterprise apps — and with new clustering tech built into the latest release, SteelCentral is ready to scale up to meet your business needs.

We spoke to  Amena Siddiqi, Director of Product Marketing at Riverbed, to find out what sets SteelCentral apart from the competition.

Little containers everywhere

Modern-day enterprise applications are often composable — that is, they're built from individual containers, and orchestrated by platforms like Kubernetes across local and cloud environments. Individual containers and the servers they run on can be spun up or down elastically as needed. This provides great flexibility and efficient use of resources, but it also makes for applications that are incredibly complex, with thousands of individual components.

In practice, this means that what looks to the user like a single transaction — paying for an item in a shopping cart, say — can actually involve literally millions of nesting method calls behind the scenes. "There are many components where something might go wrong in the course of that transaction execution," says Siddiqi. "So when you’re troubleshooting transactions, you need to be able to see into all of those nodes, because the problem could be anywhere."

The big data problem

But the reality is that most APM tools simply aren't equipped to handle the amount of data generated by monitoring so many components and transactions. These products are found wanting when they confront the three Vs of big data:

  • The volume of data involved
  • The velocity at which it flows through the application
  • The variance of metadata in individual transactions

In our shopping cart example, the items in the shopping cart and the total amount the customer would pay for them would be metadata — and if there was a problem with it you'd need to know if the problem was with a $10 cart or a $10,000 cart!

Figure 1. APM's big data problem

The typical APM vendor uses various tricks to get around these limitations. "Because they can’t scale to those tens of thousands of components, they might instrument only parts of the application," says Siddiqi. "Or they capture data based on triggers, only monitoring components if a trigger condition is met. Or they might sample what they deem to be representative transactions."

The problem with these techniques is that they produce an incomplete picture. The sampled transactions might miss the problem you're looking for. If components are instrumented selectively, often the deep layers of the application where the root problems lie are neglected, and figuring that out can take time and effort.

"If an administrator decides that they want to see what's going on with specific pieces of code," says Siddiqi, "they have to pick and choose. In order to instrument the components they want, they have to remove instrumentation elsewhere. They can't do everything because their vendor just can’t scale — they have to do this constant tradeoff between data quality and scale. And to decide which components you're interested in requires a very intimate knowledge of the code."

SteelCentral to the rescue

Now here's the good news: You don't have to make this tradeoff. With Riverbed, there's no sampling. With our proprietary data compression technology, we’ve got the smarts. We can capture everything in spite of all the volume, velocity, and variety. We just collect all the data. This is the big data approach to application performance monitoring.

Riverbed's patented technology has always delivered complete visibility at unmatched scale. And now the latest version of SteelCentral features a new clustered architecture that delivers a 10x increase in its ability to scale:

  • A single analysis server can support tens of thousands of agents capturing many billions of transactions per day, while capturing and storing every app transaction, along with system metrics at one-second intervals, with all of the relevant metadata, down to the deepest levels of the call stack.
  • The solution is benchmarked at over 20,000 and able to progressively scale beyond 100,000 instances per analysis engine instance: the highest level in the industry. Through application of proprietary data compression technology, it captures every single transaction execution.

Figure 2 illustrates the balance Riverbed brings between scale and data quality.

Crunching the numbers

With the incredibly deep store of data Riverbed's tools produce, you can perform deep analytics and derive relevant insights. For instance, Figure 3 is a performance graph generated by SteelCentral, which connects each individual transaction with the methods or SQL that's impacting it the most.

By abstracting away all the intermediate tiers that the application traverses and drilling down from the transaction type consuming the most time straight to the piece of code that needs to be optimized, you can focus development efforts on projects that have the biggest value to the business.

Figure 3. Performance graph

You can't get this depth of insight without the data that only Riverbed can provide. "You can have all the smart algorithms in the world, but they are only as good as the data you apply them to. Without a complete data set you can't get a broad enough perspective or deep enough insight," says Siddiqi. "We’ve got the data and the analytics."

APM at the scale of business

Being able to run APM at scale isn't just about bragging rights for the best technical specs. It provides real-world business benefits:

  • Alignment with business growth: Your ability to monitor keeps pace as your infrastructure grows in scale and complexity
  • Never miss a performance problem: No more blind spots for troubleshooting or business analysis
  • Simplified APM deployment: Because your whole system is monitored, you don't need to configure or maintain monitoring infrastructure to focus on hot spots

The end results?

  • Faster mean time to resolution
  • Proactively resolved production problems
  • Proper priority given to high-value efforts

"We leave you with a complete data set — no blind spots, no sampling," says Siddiqi. "This is the big data story. This is the huge differentiator that we have versus our competitors."

Find out more by visiting the Reinventing APM webpage and download our free trial.