The Automated Advertising Industry

Hyper-Speed, Hyper-Scale Advertising Automation Requires Hyper-Fast Hardware

August 22nd, 2013   |   by fbadmin

8-fast-hardwareAdvertising is following similar trends that we’ve seen in the financial industry, where trades are now executed in nanoseconds.

Years ago, we could process ad transactions in 1-2 seconds. Last year, it was 300 milliseconds. Now, it’s 30 milliseconds. Next year, it may be 3 milliseconds.

And we’re not too far away from it being nanoseconds and microseconds.

Why?

The answer is simple: devices, Internet access, Web sites and mobile apps are getting faster. And ad decisions need to happen faster than a Web site, or mobile app, loading.

The impact of this is that hyper-speed, hyper-scale advertising automation/decisioning platforms now require hardware that’s faster than hardware required to scale Web sites or mobile applications.

At Rubicon Project, we’ve engineered the world’s first real-time cloud computing technology. It’s currently comprised of 25,000 + CPUs and capable of moving 45 gigabits of data per second. And it’s processing 2 million transactions per second (that’s 350,000 in the time it takes you to blink an eye) and 3 trillion per month.

How does that stack up to big data / hyper processing platforms?

big-data

To support our real-time cloud, we’re custom engineering hardware, specifically thinking about hardware that’s designed and architected for real-time decisioning in milliseconds today, and microseconds in the future.

chart2The trend is staggering here – the transactions themselves are becoming more complex (i.e. more decisions per impression, plus more bidders), and the time allowed to process them is rapidly decreasing.

Not too long ago, the Rubicon Project had 300 milliseconds to process 1 bid for every impression, and there was a 1:1 ratio of bids to impression.

Today, due to the explosion of possible bidders and types of automated decisions/buys, there are over 100+ DSPs (i.e. 100+ possible bids + 300+ static bids from ad networks + direct orders).

And now we have only 30 milliseconds to process hundreds of bids resulting in 197,000 possible decision permutations in 1/10th the time.

The graph above illustrates the trend.

Engineering hardware is quickly going from being a competitive feature for speed and performance to a requirement for operating an automation platform and exchange at scale.