Big Themes For Data Engineering In 2019

Ternary Data heavily focuses on data engineering. So, it makes sense that we are keenly interested where things are heading for the field in 2019. Instead of doing the typical new year ritual of making crazy predictions, we consider several big themes for data engineering in 2019.

TL;DR

• Data will still be hard for most companies

• Serverless will gain stronger adoption

• Containerization will accelerate multi/hybrid cloud

• DataOps will become more widely adopted

• Data lineage will grow in importance

• Ethics issues from data science will become a data engineering problem

Data will still be hard for most companies

Despite the hype of “AI everywhere” and “data is the new oil”, let’s get real. Data is hard. The dirty secret is that most data initiatives fail. Gartner analyst Nick Heudecker sets failure rate for big data projects at “closer to 85%”.

The reasons vary, but the theme for failure is mostly the same. Chief among these reasons are lack of executive and organizational buy-in, misaligned expectations, weak data culture, lack of resources, and the talent gap. Data is hard because change management is hard. Fundamentally change management is a people problem, not a technology problem. Therefore, we expect most companies to be challenged with data well past 2019.

There are ways to win with data. We advise our clients to get back to basics with data, build a solid foundation, and start small.  We wrote some tips on starting successful data projects, which you can find here.

Serverless will gain stronger adoption

Serverless is rightfully red hot. The benefits of serverless are very clear - less undifferentiated heavy lifting and quicker time to value. The role of the data engineer is changing, from babysitting on prem Hadoop and Spark clusters to managing a plethora of interconnecting services in the cloud. Allowing data engineers to focus on code, DataOps, and integrating with data scientists is game changing and adds a ton of value.

The big cloud providers are transforming toward serverless offerings. AWS and Google Cloud have both been at the forefront of serverless for over a decade (notably, AWS launched S3 in 2006, Google Cloud launched App Engine in 2008). Azure is keeping pace. Where the big clouds go, so goes the future of cloud infrastructure. Although clouds will still continue offering managed solutions like AWS EMR and GCP Dataproc, the future direction is clearly toward serverless.

If your company is still on the fence about serverless (or managed services), now is the time to experiment. Quick smoke tests and proof of concepts are extremely cheap and easy to deploy. Get started today.

Containerization will accelerate the move to multi/hybrid cloud

For the last several years, IT has basically operated in two worlds - public cloud on AWS and on-prem. AWS’s early start in public cloud gave it a huge first-mover advantage. We have a running joke at Ternary that “AWS is the new on-prem”, meaning it’s the strong, incumbent platform.

Expect the power dynamics to shift. Kubernetes and Docker have become the de-facto standard services for containerization, making it easy to “write once, deploy anywhere”. AWS, Azure, and Google Cloud are all heavily investing in managed Kubernetes services. We believe containerization is a central strategy to Azure’s and Google Cloud’s success against AWS.

What does this mean for end-users? Gone are the days of special snowflake infrastructure. Companies using containerized applications now have the liberty to easily migrate their apps between on-prem and cloud (hybrid cloud) and between clouds (multi-cloud). Expect to see more companies using a mix of hybrid and multi cloud. Managed services from AWS such as Outposts and EKS, Azure’s AKS, and Google Cloud’s GKE will make Kubernetes a no-brainer for easily managing application clusters.

DataOps comes into its own

Compared with the adoption of continuous delivery and DevOps in application development, the time to value with data moves at a slow crawl. Making a change to a database field can take weeks or months. Such a painfully slow cadence is unacceptable in today’s fast paced, data-driven world. Time to value is a key competitive advantage for any company.

As data engineering adopts its own version of DevOps - DataOps - companies will see an improved time to value with their data. DataOps applies the disciplines of agile, continuous delivery, DevOps, and statistical process control to data. The result is that data becomes more reliable, consistent, high quality, and quickly available to downstream consumers - analysts, data scientists, automation, and models.

Data lineage will grow in importance

Questions about data commonly revolve around what and where.

“What is the origin of this data?”

“What does this field mean?”

“Where is data for X?”

Answering these questions is surprisingly difficult. Think of your data pipeline as a supply chain of physical goods. Raw data comes in, processed into a variety of outputs, and consumed. In physical supply chains, assets are tracked as they move from raw materials to finished goods. Everything is tracked as it moves through the supply chain.

Similarly, data lineage tracks the “what” and “where” of data. Data is traced from inception to end-use, and gives full visibility into where raw data ingredients are consumed. Another benefit is the ability to quickly identify data errors when they arise, and be able to address the root cause of the error. Data lineage enables better DataOps.

A few things will push data lineage to greater importance for data engineers.

First - and most obvious - as data grows in size and complexity, data will morph into various downstream processes. The result are data stockpiles. Similar to how physical supply chains track raw materials through their finished goods, tracking data assets from their raw input to end-uses will help companies keep track of their data inventory.

Second, GDPR and similar data protections now make data asset tracking a compliance necessity. Companies are now legally required to find records - and their variants - and promptly handle this data in accordance with particular laws. Data lineage is a necessary component for tracking data assets in a company’s data supply chain.

Lastly, good machine learning requires quality features. Fundamentally, machine learning models are intertwined with their training sets; each model is a soup of features and data. If the data change, applying the same algorithm will yield a different model. Data lineage allows you to know what data sources were used for what features.

Ethics issues from data science will become a data engineering problem

If 2018 was about AI and data becoming more tightly interwoven in the fabric of everyday life, it was also a very tough year. The public beca. Facebook infamously suffered from self-inflicted problems due to inappropriate data collection and sharing. China came under scrutiny for its use of AI in controlling its population. GDPR went into effect. A good summary of AI’s problems in 2018 is in the Quartz article, “This year the world woke up to the problems with AI everywhere”. Needless to say, 2018 was a doozy for the data world.

In the above article, Rumman Chowdhury, Accenture’s lead for responsible AI says “We were finding that 25% of companies were having to do a complete overhaul of their system at least once a month because of inconsistent outcomes or bias, or they were just unsure. I’m calling that ethical debt”. 2019 is the year when data engineers help solve the “ethical debt” problem in data and models.

As a side note, the IEEE is defining a new initiative called P7000, or “The IEEE Global Initiative on Ethics of Autonomous and Intelligent Systems”. We’ve been involved in discussions with this group, and are definitely excited about the direction things are going. If the issue of “ethical debt” is not addressed, we predict tough times ahead for AI and data. Our suggestion - get involved. Shape the future.

In conclusion

2019 will be an exciting year where data engineering continues to have an outsized impact. The field of data engineering is poised to create a lot of downstream value for data scientists and business in general.