Contextin Powers 10 Billion Real-Time Pricing Decisions Per Day Using the Aerospike NoSQL Database and Key-Value Store

Contextin aims to change the rules. Most conventional demand-side platforms rely on pre-determined audience criteria. By contrast, the Contextin performance advertising platform relies on an automated discovery engine. The Contextin technology goes beyond standard variables, looking at hundreds of deeper attributes in order to find the granular signals that actually influence performance, and applying that learning to every impression served. Contextin is able to dynamically infer data correlations among these non-traditional attributes on an impression-by-impression basis, and it then uses real-time bidding (RTB) to price and buy placements accordingly. The results are dramatically higher ad performance; optimized costs per mille, action or click; and the ability to capture previously untapped audiences. Discovering the best placements and the correct pricing for advertisers’ campaigns is a huge task, and the Contextin platform uses complex algorithms and pattern recognition to analyze more than 10 billion ad impressions every day. To ensure the 100% availability and real-time delivery of this data, Contextin relies on the Aerospike NoSQL database.

Contextin’s unique algorithmic approach to campaign performance optimization works by analyzing hundreds of granular variables—including page characteristics, user engagement data, and semantics—on an impression-by-impression basis and then extrapolating its learning for each campaign within the context of the specific performance and budget parameters. This enables Contextin to assess bid price and identify the type of impression most likely to get results. With massive sets of proprietary data at the core of Contextin’s platform, the company recognized the need for a powerful NoSQL database that could manage and process vast sets of information without slowing RTB response times. To support its early production platform, Contextin integrated an open source distributed database to which the company was contributing code. However, as its business grew, Contextin began to evaluate NoSQL database options that would offer greater performance. “We need to be able to hit a throughput of about 200,000 to 300,000 queries per second with response times of under 50 milliseconds for all the processing related to each query,” Mr. Naveh explains. “This is a very high load requirement, and naturally we can’t afford to have queries take a lot of time.” Query time and availability became the stumbling blocks for many of the NoSQL databases evaluated. While many of these solutions were capable of working with significant amounts of data, few were equipped to consistently provide the millisecond response times required in the online advertising industry.

Contextin became aware of the Aerospike real-time NoSQL database and decided to conduct a proof of concept to evaluate Aerospike’s performance. The company saw the results it needed. Aerospike demonstrated sub-millisecond latency and the ability to process queries consistently within 5 milliseconds. Moreover, it was able to horizontally scale to 100,000 QPS per each server in the cluster. Finally, as Aerospike processed large volumes of data, it did not slow or stall, thereby ensuring 100% availability. Further cementing Contextin’s decision were Aerospike’s cross data center replication functionality and responsive, knowledgeable technical support. Following the proof of concept and decision to adopt Aerospike, Contextin was able to move into full production in less than two months. Contextin now has an Aerospike database with cross data center replication deployed in each of four data centers: two on the East Coast, one on the West Coast, and another in Europe. Says Mr. Naveh, “Not only was the database performance significantly better than anything else we’ve used, the support of the technical team also has significantly reduced the amount of time we’ve had to spend to get operations up and running.” Contextin uses the Aerospike database to respond to queries by combining real-time data with information from the other data management and business intelligence systems it has deployed. These include terabytes of unduplicated data stored in its Apache Hadoop distributed file system, MySQL relational database management system, and Qlikview data analysis platform. Aerospike’s ability to conduct read and write transactions within 5 milliseconds helps Contextin power more than 10 billion pricing decisions per day. In addition to providing reliable real-time performance, the Aerospike database automatically manages itself, minimizing the demands on Contextin’s IT team. Currently, one internal IT administrator is able to easily manage all of the company’s Aerospike deployments. Equally important is the technical support Contextin receives from Aerospike.

• Contextin uses the Aerospike database to respond to queries by combining real-time data with information from the other data management and business intelligence systems it has deployed.
• Aerospike’s ability to conduct read and write transactions within 5 milliseconds helps Contextin power more than 10 billion pricing decisions per day.
• In addition to providing reliable real-time performance, the Aerospike database automatically manages itself, minimizing the demands on Contextin’s IT team.
• Currently, one internal IT administrator is able to easily manage all of the company’s Aerospike deployments.
• Equally important is the technical support Contextin receives from Aerospike.

• Aerospike demonstrated sub-millisecond latency and the ability to process queries consistently within 5 milliseconds.
• Aerospike was able to horizontally scale to 100,000 QPS per each server in the cluster.
• Contextin was able to move into full production in less than two months.