When 10 Milliseconds Matter: Choosing the Right NoSQL Database for Mobile Advertising (Article mentioning Aerospike - May 18, 2016)


#1

The article below appeared on Adotas here on May 18, 2016.


Mobile ad spend will account for 72% of total digital ad spend and a $65B global market by 2019 (up from $28B and 49% last year). More than any other industry, its viability hinges on the ability to make critical decisions in just 10 milliseconds.

Below is an Adotas Q&A with Kai Sung, CTO of Manage, a company that sells in-app ads, and Brian Bulkowski, Founder, CTO & Product, Aerospike, a NoSQL database company, that explores database selection, management and what it takes to succeed in the ever-changing world of digital advertising.

Q: How should ad tech companies go about the process of selecting a database?

A:

Brian: The right database can give an ad tech company real competitive advantage, so it’s an important decision. Listing your requirements is a great place to start. Millisecond response times, billions of objects, consistent performance, low latency and scalability are priorities for many ad tech companies. It’s all about which database can keep up with your business goals. And databases aren’t a one-size-fits-all solution any more, you need to pick the right database for the right job.

Kai: Once you narrow it down to two or three options that meet your requirements, do a Proof of Concept test. Installing the technology and simulating workloads is the best way to find out if a database can support your demand. Its also a good idea to run benchmarks to evaluate things like memory usage and read performance CPU.

Q: How have the new demands of the real time bidding era—the need to process huge amounts of bid requests in milliseconds—changed this process?

A:

Brian: One of the most important aspects of ad tech is catching the right audience at the right time. Companies like Manage use real-time bidding exchanges to identify the best inventory to bid on behalf of their advertisers, and then they optimize campaigns toward each advertiser’s goals. These bid requests come in at hundreds of thousands per second, 24 hours a day, all year long. To make accurate predictions for every bid, it’s crucial for the database to have a predictable millisecond response time, and to achieve that with just a few servers.

Q: What are the top three considerations ad tech companies should keep in mind when it comes to making data-based decisions?

A:

Brian:

  • First: Cost-effectiveness of data. If you work to reduce the cost of your front-edge data, you can retain more data, engage more audiences and trial new algorithms. Every time you push down the cost of your data, you gain competitive edge.
  • Second: Actionableness. Keeping your eye on actionable data is always key. When you know how often a particular algorithm is effective, leading to not only a won bid but profit on that bid, you’ll continually improve.
  • Third: Predictability. It’s hard to write code that makes a decision when you don’t know how long it’ll take to execute the computation. Predictability means measuring 90% and 95% latencies.

Kai: My top three?

  • First, quality. Being able to extract the right signals from the noise, and being able to filter out fraud.
  • Then, speed. Not only being able to respond within milliseconds to a bid request, but also having fast data pipelines to ensure the data we’re using is fresh and to enable a proactive reaction on our part.
  • And also, scalability. Being able to process upwards of 40 billion bid requests per day, and scale up and down as traffic spikes throughout the day.

Q: How does header bidding help publishers streamline processes and increase revenue?

A:

Kai: For the publisher, header bidding frees them from having to flight multiple ad lines, and instead just flight one ad line. It also allows publishers to market their inventory to a pool of multiple interested buyers in real-time. Because these buyers are able to assign more precise value to higher quality inventory, this can significantly increase a publisher’s revenue.

For DSPs, header bidding creates more equal opportunities to bid on impressions, ones we might not otherwise see under the waterfall model. We believe that every impression has some value ascribed to it, and it’s our job to help determine what that value is and make the appropriate bid on the advertiser’s behalf.

Q: What’s the biggest challenge facing the digital advertising industry right now?

A:

Brian: Managing and leveraging data sets that are just enormous and continuously growing—and the technology chain both for storage and for implementing new technologies like machine learning. Ad tech companies need to store enough data in real time and put the data to use. A solution to this is separating the analytics tier from the real-time tier, and choosing the correct application environment and database for each.

Kai: The advertising landscape is becoming more fragmented in terms of inventory. Inventory is growing exponentially, and it’s becoming more of a challenge to glean what’s high quality versus not. Also, consumers are using multiple devices, making cross-device matching and targeting more and more critical to be able to paint a holistic view of the user.

Q: What is behind the rise of ad blocking, and where will this lead?

A:

Brian: Well to the average Internet user, ad blocking seem helpful and convenient. Many studies (i.e. eMarketer, MarketingLand) show significant adblock rates in valuable younger customer segments. But behind the scenes, the ads people are trying to avoid are the foundation of a diverse Internet. The ad industry typically faces major new challenges every year, from click-fraud to viewability, and now adblocking—and eventually solves every one. Today, all publishers have the choice of paying adblock companies for relief, building a business model even with some ads blocked, relying more on mobile applications, building unblockable in-line insertion technology, and creating a subscription path (stand-alone or aggregated)— or a blend of those strategies. Navigating that landscape is becoming part of what a publisher must do.

Kai: At the end of the day, consumers just want relevance and transparency when it comes to advertising. Our goal is to maximize ad relevance using the wealth of data and signals we have access to, and to also give consumers an out if they don’t want to see ads.

Q: Where do you see the digital advertising market heading in the next 5 years?

A:

Kai: Publishers and formats will be expanding—we’re seeing more interest in new mediums like live streaming, connected TV and even VR. We also anticipate the continued convergence of performance and brand marketing—just look at the trend of major app companies partnering with household name celebrities to build brand awareness. Manage is positioning itself to serve both of these marketer needs.


Brian Bulkowski, Founder, CTO & Product, Aerospike

Brian is a founder of Aerospike, CTO & Product, networking whiz, innovator and high performance expert. Brian started as a Lead Engineer at Novell, and then Chief Architect of Cable Solutions at Liberate – where he built a high-performance, embedded networking stack, as well as the high scale broadcast server infrastructure. As Director of Performance at Aggregate Knowledge, Brian had direct experience with the scaling limitations of sharded MySQL systems. Brian’s desire to eliminate those limitations lead to the creation of Aerospike. When he’s not busy creating stuff with plywood or a welding torch, Brian plays cello in a band called Rosincoven. He also writes about cuisine for the San Jose Metro.

Kai Sung, CTO, Manage

Kai co-founded and was CEO/CTO for AppBank.com, a Top 5 Facebook app developer according to AppData that reached over 30 million monthly active users in 2010. Prior to that, Kai held senior engineering positions at Voltage Security, Sun Microsystems, and HP/Compaq. Kai holds a Bachelors degree in Computer Science from UCLA, and a Masters degree in Computer Science from Stanford. Kai is an avid marathon runner and outdoorsman.