The ‘Ops’ concept takes hold in enterprise technology shops, but so do new headaches

DevOps — which fosters greater collaboration and automation in software delivery — is only the beginning of a new phase of technology management. Now, we are seeing many spinoffs — DataOps, Machine Learning Operations (MLOps), ModelOps — and other Ops that seek to add speed, reliability, and collaboration to the delivery of software and data across enterprise channels. There is even a DataOps Manifesto, which bears a striking resemblance to the Agile Manifesto crafted back in 2001.  

However, none of this stuff is going to happen overnight. Or even within a few months. As with any promising technology overhaul, a rethinking of processes and culture is essential. 

Where does that leave IT managers and professionals? How should they proceed with all these Ops promising smoother and more responsive service delivery?  “A key element of preparation is to ask the important questions about existing processes, both formal and informal,” says Alice McClure, director of artificial intelligence and analytics for SAS. “This helps identify where to focus first, what needs to be updated and where bottlenecks exist.” 

DataOps, for one, “provides an agile approach to data access, quality, preparation, and governance — the entire data lifecycle, from preparation to reporting,” says McClure. “It enables greater reliability, speed and collaboration in your efforts to operationalize data and analytic workflows. ModelOps is becoming a must-have methodology for implementing scalable predictive analytics. It’s all about getting analytics into production – iteratively moving models through the analytics life cycle quickly while ensuring quality and enabling ongoing monitoring and governance of models over time.”

It’s all about bringing together automation and architecture, advises Amar Arsikere, CTO and co-founder at InfoWorks. “Deploying a system that automates data, metadata, and workloads operation and orchestration, versus hand-coded, manual operations that take time, money, and specialized resources.” 

 xOps approaches are becoming a necessity as manual-adverse applications such as artificial intelligence and machine learning come to the fore. “Addressing these challenges is often an afterthought and eventually falls on DevOps and IT teams,” says Rahul Pradhan, VP of product and strategy for cloud platforms for Couchbase. Emerging priorities such as continuous integration and continuous delivery, automation and real-time monitoring are putting a strain on these teams, he adds. “Not only are these teams being asked to do more, they are also being asked to be broader and full-stack. This highlights the need to eliminate operational low-value tasks like managing infrastructure and databases.”

Most operations “are heavily scripted or automated, but real success is achieved when the entire process is automated from start to finish,” agrees Patrick McFadin, VP of developer relations at DataStax. “This includes the day-two operations, such as scaling. xOps can follow a similar path that site reliability engineers take for training and preparation, since they deal with the same issues in cloud-native applications.”

Contrary to popular belief, having a successful xOps effort doesn’t mean enterprises can reduce their IT staffing levels — if anything, it means they need to step up their recruiting and retention games. IT talent shortages “can significantly hinder xOps initiatives,” says Pradhan. “Direct more effort towards developer retention. By taking proactive steps to keep developers engaged and satisfied, digital transformation burnout can be avoided.”

There’s another key element in xOps success: time to deploy and overcoming stale corporate cultures. A new ModelOps or DataOps methodology “can’t be implemented and built in a day,” Pradhan points out. “It takes time to transform processes. Involving the right teams at the beginning of a project is critical and should include crafting quantifiable outcomes and a clear understanding of roles.” 

The challenge is “shifting teams’ mindsets to be organized around the business transformation goals and outcomes,” says Arsikere. “Rethinking deployment by automating end-to-end processes instead of relying on manual hand-coding, or disparate point solutions.”

That’s where Ops methodologies “can help simplify things, with  to drive business value, while ensuring the best customer experience,” Pradhan urges. He urges a composable approach — similar to a Lego building-block strategy — “to help ease tension that can occur as xOps capabilities and digital transformation strategies are being built. The same blocks and strategy can be used again and again.”

In addition, it’s time to bring application and data infrastructure development and deployment under one roof, says McFadin. “Don’t hang on to old methodologies,” he says. “I often see enterprises separating application and data infrastructure with different methods and standards. Committing to a single path for both code and data can open up a lot of capability. That means finding ways to make the data portion of the application stack cloud native.”

Embracing cloud-native for data “separates the teams that move fast from those that don’t,” says McFadin. “That means utilizing everything available in the Kubernetes ecosystem to their advantage. From CI/CD to observability, the goal is to create repeatable and trusted systems. DevOps has had an early lead with projects that address different problems. MLOps and DataOps are now quickly catching up with new and emerging projects.”