Ad servers need to be fast—so fast, that users do not notice any lag of external content being loaded. Typical ad servers select, serve, and track ad impressions in milliseconds. They offer seamless experience with lightning fast response times, even when handling thousands of ad requests and event-tracking operations every second.
It is crucial that you optimize your ad server for excellent throughput (number of requests per second). Make sure your system architecture is well planned and agile. Use small components that gel together for fine performance. The advantage is, even if you find the system to be slower than expected, you can identify and replace the module that is not fast enough. That helps you remain efficient and quick, rather than having to rework or replace huge modules.
Test it for performance by bombarding your application with massive traffic. When it stops being responsive or starts throwing exceptions, verify the logs, identify problem areas, and then fix them.
Ad servers can be optimized by using highly concurrent event-driven technologies. Let us consider a few like Akka, ZeroMQ, and Spray. Using such technologies helps to keep the cost of ownership low.
|
Akka (http://akka.io) is a toolkit and runtime for building highly concurrent, distributed, and resilient message-driven applications.
|
ZeroMQ (http://zeromq.org) is a high-performance asynchronous messaging library. It comprises high-speed asynchronous I/O engines in a tiny library and is backed by a large and active open-source community. |
Spray (http://spray.io) is an open-source toolkit for building REST/HTTP-based integration layers on top of Scala and Akka. Being asynchronous, actor-based, fast, lightweight, modular, and testable, it is a great way to connect your Scala applications to the world.
|
Scalability
Scalability is another important aspect that affects performance and throughput. Scalability implies the system’s ability to avail new resources, so that the application runs smooth. Adding server machines for the application is horizontal scaling; and upgrading existing server machines is vertical scaling. Horizontal scaling may create problems for applications with server affinity; but vertical scaling works well independent of application design.
Being concurrent in nature, Akka, ZeroMQ, and Spray enable horizontal scaling down of ad server solutions to achieve higher throughput with similar server configurations.
