Skip to content

Targeting data at the speed of an ad auction.

Proprietary targeting data is useful only if it reaches the auction in time. We built a serving layer to make that data available under demanding response-time constraints.

Project:AdDaptive IntelligenceA past project. The story describes the work delivered and the technology used at the time.

The problem

AdDaptive wanted to make its proprietary business data available inside advertising exchanges. The system had to serve that data at low latency and move it from AWS to the infrastructure hosting the bidder.

A messenger robot delivering targeting data directly into an advertising-auction service.

What we built

We prototyped and benchmarked combinations of databases and HTTP frameworks, then built the serving layer with Aerospike and Netty.

  • Compared prototypes for performance, development effort, and maintenance.
  • Implemented an HTTP serving layer in Java with Netty.
  • Loaded targeting data from AWS into Aerospike in the bidder’s data center.

What changed

A working data service with a manageable footprint.

The delivered system handled the workload at low latency on a small set of commodity servers. The Java implementation also fit the team’s existing skills, making subsequent changes easier to support.

The technical decisions

Benchmark before choosing.

We evaluated candidate approaches against the latency requirement instead of selecting a familiar database first and hoping it would fit.

Moving the data mattered too.

Aerospike’s data model and loading process supported bringing data from AWS into the serving environment. The ingestion path was part of the design, not a separate afterthought.

Let’s talk

Have a similar problem?

A rough idea is enough. We’ll work out the next step together.

Tell us what you need