CASE STUDY: Rubicon Project Automates Real Time Business Incident Detection with Anodot
About the Company
One of the largest ad exchanges in the world, Rubicon Project uses proprietary computing systems to automate the buying and selling of advertising. The company employs more than 700 people and provides real time ad auction services for their partners in various industries. More than 90 percent of people browsing the internet will see an ad that goes through the Rubicon exchange; each day, Rubicon’s systems process more than twice as many transactions as the Nasdaq stock exchange.
Massive Amounts of Data Need Real Time Incident Detection
As one of the most successful global advertising technology companies, Rubicon Project processes trillions of transactions each month in real-time auctions that each occur within 40 milliseconds. The company receives more than 13 trillion bid requests per month, that are handled in its seven global data centers, housing more than 55,000 CPUs. Rubicon’s life blood are its data centers, which analyze billions of data points real time.
Rubicon’s Tech Ops team monitors each of its CPUs, racks and data centers. This team, however, cannot monitor more complex aspects of business and trends, and certainly not in real time. For example, Rubicon needs real-time insight if a large institutional buyer moves off its normal transaction trend by any percentage in one of the global data centers at any hour of the day or night in order to understand if something is wrong. Deviations can have a devastating effect on the exchange if there is a lag in addressing it with the customer.
Currently, Rubicon works with many demand-side platforms (DSPs) across its global data centers in different time zones. Along the bid stream, there are many potential areas for communication or technical breakdown, which would prevent the bid from going into the auction, and negatively affect overall bid health.
Anodot Simplifies Real-Time Analytics
With 13 trillion monthly bid requests, 55,000 CPUs and 7 data centers, Rubicon’s needs have grown beyond the scope of what a human or group of humans can feasibly monitor and control. With Anodot, running on AWS, leveraging S3, EMR and other components, Rubicon can easily monitor all of their data in real time to aid in the creation of a fair and healthy ad marketplace.
The company was already using Graphite for its monitoring, so it simply pulled Graphite data into Anodot. They immediately benefitted from streamlining and automating the data analytics.
“The teams using the tools are more business-focused and straddling technology, so we needed something simple, not code heavy,” said Rich Galan, Director Analytics at Rubicon Project. “Anodot’s tools were so simple to integrate and we’ve had no trouble learning and using them.”
Previous monitoring tools that Rubicon used required the company to manually set thresholds to generate alerts. Anodot takes the solution to the next level, by learning the normal behavior of Rubicon’s data and determining seasonality for each metric, unlike other solutions which cannot account for seasonal trends.
During the Proof of Concept stage, Rubicon experimented with another so-called anomaly detection tool, and found that the amount of programming and technical understanding needed for its setup was not a good fit.
“We generally prefer to build all our tools internally, but after working with Anodot, our Chief Data Scientist estimated that it would have taken at least six of our data scientists and engineers more than a year to build something of this caliber,” Galan said. “So it was a no-brainer for us to jump on board and take it. Our head of tech ops told us that he’s been searching for years for an automated data analysis solution like this.”
With the new real-time analysis from Anodot, Rubicon will notice if a specific DSP stops responding to bids over 15 minute increments. Rubicon uses the Anodot alerts and correlations to determine if they need to reach out to their partner to fix an issue. Sometimes, the partner is already aware of the issue, but other times, Rubicon found, the partner had no idea there was a technical problem and was happy to be informed so that they could fix it. “We’re not just helping our business, we’re helping our partner’s business. That’s the level of service that we want to provide. We are the biggest exchange and we want our partners to understand that we’re watching things very closely not just for us, but for them as well,” Galan said.
Anodot for the Long Haul
Rubicon expects to have dozens of people from various business and analytics teams using the Anodot system, which will be tracking tens of thousands of metrics across the business.
The ease of use of Anodot business incident detection, together with its automatic correlations makes it straightforward for account managers to quickly see the problems that need to be addressed with their clients immediately. If the Anodot system detects a fall in revenue, it will correlate relevant other metrics; for example, blocks could be up, timeouts could be up, bid responses and pricing could be down. With these correlated metrics, Rubicon can quickly pinpoint where to investigate.
“The Anodot solution has so many useful tools, Our team will benefit from the real time alerting, the simplicity, the seasonality and the correlations,” Galan concluded.
“We generally prefer to build all our tools internally, but after working with Anodot, our Chief Data Scientist estimated that it would have taken at least six of our data scientists and engineers more than a year to build something of this caliber.”
– Rich Galan, Director Analytics