Case Studies Federated Media Publishing Powers Third Largest Ad Network With Aerospike
Edit This Case Study Record

Federated Media Publishing Powers Third Largest Ad Network With Aerospike

Analytics & Modeling - Real Time Analytics
Application Infrastructure & Middleware - Database Management & Storage
Infrastructure as a Service (IaaS) - Cloud Storage Services
Business Operation
Sales & Marketing
Real-Time Location System (RTLS)
Software Design & Engineering Services
System Integration
The Federated Media Publisher Network platform powers ad delivery through a distributed architecture consisting of a main datacenter that communicates with cloud-based advertising “pods” to work with demand-side platforms, ad exchanges and ad networks for real-time bidding (RTB). Federated Media Publishing also guarantees brand safety for advertisers by looking at page-level context to ensure that ads are served on brand-safe pages, as well as providing targeting through page-level site categorization. Delivering these functions in real-time is an integral part of the company’s new focus on programmatic buying, which automates the placement process by instantaneously selecting who to serve impressions to based on data an advertiser thinks is pertinent to the campaign. Federated Media Publishing quickly determined that relational databases could not provide the ultra-fast response times required to deliver real-time services across its platform. Instead, the company began evaluating NoSQL databases for their ability to handle high volumes of data and respond in milliseconds.
Read More
Federated Media Publishing, along with its wholly-owned subsidiary Lijit Networks, provides advertising services, audience analytics, and reader engagement tools to over 145,000 sites on the “Independent Web.” The company leverages the Aerospike real-time NoSQL database and key-value store to manage ad impressions for more than 180 million unique monthly visitors in real-time. This scale of interactions has led comScore to identify the Federated Media Publisher Network (FMPN) as the industry’s third-largest network in the display ad ecosystem. Federated Media Publishing aims to level the playing field for independent publishers who lack the resources of large media companies and major service platforms.
Read More
Federated Media Publishing narrowed its NoSQL choices to a few different vendors. The company then began a series of evaluations focused on the request response time, ease of use, and replication between the data center and advertising pods. While Aerospike demonstrated sub-millisecond latency and the ability to process queries consistently within 5 milliseconds, Federated Media Publishing’s replication requirements initially provided an obstacle. Aerospike delivered the speed and reliability they were looking for, but it only offered a ring topology for replication. Federated Media Publishing wanted a star topology, so their main data center could replicate the same data to all other servers. Aerospike responded quickly to this need, developing the star topology within six weeks. With star topology replication on hand, FMPN began a proof of concept (POC) with the Aerospike database, testing speed, scale, replication, and reliability. They discussed with the Aerospike team how the database was designed to be self-automated and self-tuning, taking servers down to see if clusters stayed up.
Read More
Federated Media Publishing has consolidated its cloud-based pods into five hosted data centers, including the company’s main data center.
Aerospike clusters at each of these data centers run Intel-based servers using the Linux operating system.
Each cluster manages 50 billion objects and holds 30GB to 40GB of data; cross data center replication across the five data centers ensures continuity.
Federated Media Publishing manages ad impressions for more than 180 million unique monthly visitors in real-time.
Aerospike demonstrated sub-millisecond latency and the ability to process queries consistently within 5 milliseconds.
Federated Media Publishing needed data distribution to happen within 125 milliseconds or less.
Download PDF Version
test test