DevOps is poised for healthy growth over the next several years, with analysts large and small weighing in with forecasts. For instance, IDC predicts the worldwide DevOps software market will reach $8 billion by 2022, up from $3.9 billion in 2017. Similarly, Grand View Research estimates the global DevOps market will hit $12.85 billion by 2025, registering a CAGR of 18.6 percent.
Below are some of the most influential trends driving growth in 2019.
DevOps Instrumentation in AI and Machine Learning will Increase Productivity Gains
Web-based forms are the nemesis of process automation. Everyone dreads manual forms, but that’s what organizations use when they want to capture data from customers, employees or partners.
AI and machine learning technologies are mature enough to reliably process voice, video, text and images. That means data-enriched forms could be filled out simply by making a phone call, taking a video, taking a picture or some other natural activity. The analog inputs from the external world are captured digitally and automatically.
Of course, the error rates in such conversions would depend on how well the AI models have been trained and how well we use the real-world input to enhance the model. DevOps-level instrumentation needs to be introduced in the pipeline. That way, either the customer or the person in-loop will be able to take these errors as inputs for enhancing the accuracy of the models.
Key Management is Key for DevOps
In 2019 and beyond, data usage will be closely monitored as GDPR becomes a reality and similar privacy laws are passed across the globe. This is when data tagging becomes critical. Therefore, tools related to data tagging and master data management will become crucial for tracking the origin of data usage.
Sensitive data will need to be encrypted in rest and in motion. Along with data encryption comes the important responsibility of managing encryption keys. Not only should the keys be stored securely, they also may need to be rotated. Key management services that offer such functionality will become commonplace in the DevOps world.
Container Adoption Increases with Data Locality Demands
Data locality is driven by countries that mandate data must reside within their geographical boundaries. Enterprises using SaaS or PaaS will use country-specific public clouds or even private clouds. As a result, critical data and applications will have to be deployed across geographies. Deployments will become more complex due to geographical expansion as well as agile development methodologies.
In response, container-based development, release and deployment models will become commonplace within the enterprise. While containers offer many advantages by creating siloed workloads, they also increase the demand for sophisticated monitoring tools and technologies that help consolidate the view of these applications.
The Compiler is the New King of the Data Center
The demise of Moore’s Law is not helping the CPU keep up with increasing workloads in the data center. To bridge the performance gap, modern hardware such as GPUs, FPGAs and ASICs will infiltrate the data center.
Developers will start using higher-level languages such as OpenCL and CUDA, and the compilers will generate code to run on the various hardware, depending on deployment-level input including workload, hardware configuration and data size. DevOps practices will have to start befriending compilers so that code can be generated just in time, very close to the point of deployment.
Technological advancements, security requirements and legislative changes will impact the professionals within this sector, which is growing at a rapid pace that could see it double within five years. As developers prepare to manage a higher volume of sophisticated data, an increasing amount of DevOps solutions will enter the market to further automate processes, accelerate deployments and address the above trends.
Giving automation the power to detect crime and enforce punishment has ramifications, even for minor infractions
In November, I broke the law. I crossed over a solid white line to make a right turn at a traffic intersection. At the time I was unaware of my violation. I was on my way to a shopping mall in an unfamiliar part of town to buy my wife a gift for her birthday. My only defense is that I was following the instructions emitted from the map app on my cellphone. It told me to make a right turn. So I did. Little did I know I was being watched.
Automation enhances law enforcement capabilities
A few weeks later, I received a citation in the mail. The citation had photos of me, my car’s license plate and the car crossing the solid white line. Turns out, there were cameras at the intersection watching all traffic activity. I was busted. I was also a bit shocked: My experience getting a traffic ticket 10 years earlier involved a real police officer writing me up in real time. Now it was as if Big Brother was indeed watching.
In addition to the eerie feeling that goes with being charged with a crime anonymously well after the fact, I also felt baffled. The traffic citation did not list the amount of the fine I had to pay. The only information it provided was that I needed to go to the traffic court website to learn the details. So I did.
I logged into the website and entered my citation number. The site had no record of the ticket, but noted that it can take up to 30 days for an automated ticket to make it into the system. I felt powerless; I was charged with a crime, liable for a fine and yet didn’t know how much my transgression was going to cost me.
I decided to call the telephone number provided on the citation for more information. I dialed the number using the same cellphone that had gotten me into trouble in the first place. I was connected to an automated answering system. After navigating through various Yes/No and “enter the proper number” prompts, I learned that the system had no record of my citation. My only option was to wait 30 days for the systems to get their collective act together.
A month later, a formal citation did indeed arrive in the mail, noting the amount of my fine ($480) and options for payment. Also, the citation provided instructions to go online and pay the ~$500 required to attend traffic school, should I so desire. (Traffic school is an option in California in which you attend a one-day class about traffic safety and your violation is kept private from insurance companies. The result is, your automobile insurance rates don’t go up. Here in Los Angeles, a lot of professional actors teach traffic school as a side job. As you can imagine, traffic school can be quite entertaining. But, I digress.)
At this point, I was completely bewildered. I’d been busted for a crime I didn’t know I committed, by machines I didn’t even know were there. The only way I could get the details about my transgression was by interacting with other machines on the internet or on the phone. And, should I decide to plead guilty and settle matters, my first, best option is to do so via automation.
Still, the technology was failing me. I needed to talk to a person, any person. So I went down to the courthouse in Santa Monica. There I found a special window on the exterior of the courthouse building for processing traffic inquiries. (One of the nice things about living in sunny Southern California is that providing service outdoors throughout the year is a viable option.)
I waited in line. Finally, I got called forward to the window—which, by the way, was a pane of blackened glass. I had no way of knowing who or what I was talking to; all I saw was my reflection in the black mirror. A voice from the speaker at the bottom of the window asked me to state my business. I did. The conversation continued. The responses from the speaker seemed brusque, if not a bit preformulated. I started to wonder, Am I talking to a machine or a person? So I asked a random question, “How much is two plus two.” The voice in the speaker responded, “Sorry I am not programmed to respond to that information.”
Now I was really thrown. Was I confronting yet another piece of automation? After a minute or two, the voice in the speaker said, “I’m just messing with you. I get this all the time. They all think I’m a machine.”
It turns out that the voice did indeed belong to a human, and the human’s name is Dante Pipkins. Dante shared an interesting fact with me: Even though my violation was caught on camera, the actual OK to create the citation was issued by a human police officer behind the scenes. Go figure.
Thus, having been enlighted about the details of enforcing traffic law via camera, I finished my business by having Dante give me a date to appear in traffic court, before a human judge, to plead my case. I knew there was no way to proclaim my innocence. I had been caught red-handed. My intention was to throw myself on the mercy of the court to get a reduced fine.
The Power of Automation
Automation is a great thing. It empowers us. It saves us time and money. Without automation, there’s no AWS, Google, Azure, Digital Ocean, Netflix, Etsy, Shopify, Facebook, Instagram, Slack, Airbnb, New York Stock Exchange … the list goes on. Automation is more than a technological feature. It’s now a way of life. And, for those of us who work in DevOps, automation is fundamental—none of us want to go back to the days of manual deployment and guesswork system monitoring, to say the least.
Yet, we do walk a tightrope. As my recent travails with a very minor aspect of the legal system demonstrate, automation is becoming a key factor in everyday law enforcement. On face value, this might seem like no big deal, but consider this: Today, charging a citizen with a crime is still subject to some sort of human intervention. There are a lot of details to consider that go beyond the current capabilities of AI and machine automation. Human judgment is still the most reliable form of judgment. Yet, it’s time-consuming and expensive. Eliminating time-consuming, expensive tasks with cheap, fast, accurate machine-driven processes is a key motivator for using automation.
As automation becomes more familiar with the activities it is processing, it becomes an expert. And, as automation achieves expert status, it replaces human activity. Just think about detecting credit card fraud; these days, practically all of it is done using automation. While we still have a long way to go in terms of totally automated processes, automation is creeping its way into law enforcement. My recent experience is proof.
Now, imagine a world in which crimes such as tax evasion, financial fraud, embezzlement, regulatory non-compliance and yes, moving traffic violations are detected, reported and, most importantly, prosecuted using automation. Let’s go further: Let’s imagine that automation becomes the default method of prosecution. Imagine that the justice system becomes so automated that the defendant is presented with an offer to plead guilty to a sentence already predetermined according to an AI algorithm. Or, the defendant can choose to opt into a human interaction but will need to bear the burden of the increased court costs, very much in the same way that an automated audio transcription service charges extra for a human to do the work.
Is such a scenario possible? Well, think about this: My interactions around the traffic violation were conducted without me ever having once had to interact with a human being. I was guilty as charged and automation proved this beyond a reasonable doubt. Not only did the technology detect my transgression, but it also connected me to my car, thus making citation possible. Hence, there’s a good argument that, when it comes to automation-enabled law enforcement, the distance from catching misbehaving drivers in the act to actually going after real crooks is just not that far a leap.
So, what does this have to do with DevOps? When you scrape away all of the technological marvels, at some point there’s going to be someone in DevOps either creating, deploying or fixing the systems that make this all possible. It’s a big deal to give automation the power and technical wherewithal to enforce “the law.” As the feasibility of putting complex, comprehensive law enforcement systems into play increases, are we in DevOps going to consider such deployments as just another day in the CI/CD pipeline, or are we going to think through the game-changing impact of what we’re about to make?
VMware has updated its Wavefront monitoring service to include support for distributed tracing along with anomaly detection capabilities enabled by machine learning algorithms.
Stela Udovicic, senior director for product marketing for Wavefront by VMware, said the addition of the distributed tracing support, now available as part of a public beta, means that a single monitoring service now can provide metrics, histograms and the tracing function required to provide a holistic view of the IT environment. Those capabilities eliminate the need to employ multiple services to achieve the same goal, she noted.
In contrast, Wavefront makes it easier to search for traces containing a specific service or application programming interface (API) to determine which ones have errors or long response time. Additionally, DevOps teams can filter the trace search to show traces containing spans from a given cluster or shard.
At the same time, VMware is adding an AI Genie capability that promises to make it easier to troubleshoot existing needs and identify future application and IT infrastructure requirements, said Udovicic. That data is collected in real time and presented in a way that allows even the most novice users to visualize using analytics tools built into the Wavefront services, added Udovicic.
Wavefront is based on an agentless architecture that in part makes it possible for the service to process millions of metrics, histograms and traces per second. That capability is especially critical in IT environments that have embraced microservices based on containers that tend to expand across an enterprise IT environment, said Udovicic. Wavefront already supports Kubernetes clusters.
In general, Udovicic noted that developers tend to embrace a variety of open source approaches to monitoring applications as they build their applications. But a SaaS application such as Wavefront makes it easier for enterprise IT organizations to apply monitoring across a broad range of applications. To facilitate that transition, Wavefront provides an OpenTracing/OpenCensus compliant solution that includes a drop-in replacement for the Zipkin and Jaeger distributed tracing software, which is at the core of many open source monitoring tools.
VMware is clearly betting that increased reliance on AI will push organizations toward relying more on cloud services. Machine learning algorithms require access to massive amounts of data before they can construct AI models capable of accurately predicting events. The cost of collecting and storing all that data, plus the expertise required to train the AI models, are beyond the capabilities of most DevOps teams.
Most enterprise IT organizations have applied monitoring tools to only their most critical applications, mainly because of cost concerns. But as applications based on microservices proliferate across the enterprise, it’s becoming apparent that all the dependencies that will exist between various application services will need to be continuously monitored. The only issue now is determining how best to achieve that goal as part of an integrated set of DevOps processes.
DevOps has quickly matured from a fairy tale to a bestseller at most large enterprises. However, we are still at the beginning of the DevOps story, with many exciting chapters still to be written. This article looks at several emerging technologies shaping DevOps today and a handful of developing trends that will shape the ecosystem in the years to come.
Today’s IT organizations increasingly are moving away from monolithic applications deployed through grand-slam releases with security slapped on after the fact. We are seeing many enterprises transitioning to more atomic applications released in continuous deployments with application security baked in from the beginning. Fueling today’s emerging DevOps is a bevy of new products focused on architectures including microservices, service mesh and function-as-a-service (FaaS); advancements in pipeline automation; and more adaptable security paradigms such as interactive application security testing (IAST) and runtime application self-protection (RASP). While these solutions herald a great step forward, they do demand that legacy applications be refactored into smaller granularity services, layered onto newer architectures and migrated to processes that allow greater automation.
OK, that’s where we sit today: Great progress creating great opportunities and great challenges. But let’s head a little further down the road to understand where today’s advancements are driving DevOps tomorrow.
There are five trends consistently appearing on the whiteboards of venture firms and forward-thinking startups that will shape DevOps in the upcoming years. At a macro-level, DevOps must become wider and smarter. As we dig layer deeper, we see five distinct patterns appear in which emerging technologies will make the release pipeline more insightful and proactive while allowing it to handle a wider spectrum of pipeline artifacts beyond just applications:
AI for DevOps: Artificial intelligence (AI) is making inroads into all facets of IT and DevOps is no exception. AI techniques soon will make the DevOps pipeline smarter, able to predict the impact and risk of deployments, spot procedural bottlenecks and identify automation shortcuts. Advancements in robotic process automation (RPA) lend themselves directly to optimizing the various handoff and automation points of the release pipeline. AI-based predictive analytics also will allow for higher fidelity operational capacity planning and pre-deployment fault prediction.
DevOps for AI: Today we are seeing a handful of new tools, commonly called MLOps, emerging on the market that allow developers and modelers to deploy new AI models rapidly and repeatably. However, these products have yet to collaborate closely with the Ops side of the enterprise. They are generally a development aid, not an operational aid—or said another way, they are all “ML” and very little “Ops.” This space will mature rapidly, however, allowing release of AI models via the DevOps ecosystem so as to include operational aspects. Conversely, tomorrow’s DevOps pipeline will also allow AI models that have been learning from real-world data to be extracted from production and moved back into the development environment.
DBOps: Today, new databases, schemas, indexes, triggers and stored procedures typically are deployed manually by DBAs outside the DevOps pipeline, with their operational impact being provided by best-guess capacity planning. This glaring gap soon will be closed by products that allow databases and their associated artifacts to be released through the deployment pipeline along with the applications with which they are coupled. This will not just encompass traditional relational databases, of course, but also support non-structured databases and object stores. The prime objective being that applications and their supporting data repositories will be developed, tested and deployed via the same automated pipeline as applications without the need for specialized database handholding.
DataOps: When new apps are deployed, the associated analytics and visualizations are typically brought online by data scientists after the fact through manual deployments. Looking forward, expect data analytics to be introduced into the DevOps pipeline soon after the requirements for the release are defined. Much like DBOps above, the goal is to eliminate specialized manual promotion. But beyond just synchronous deployment, DataOps will allow data governance, data loss prevention (DLP), data lifecycle management and copy data management features to be integrated during each new deployment.
Blockchain for DevOps: Today’s and tomorrow’s DevOps evolutions will increase the velocity, variety and volume of pipeline traffic dramatically. While this sounds promising, it undoubtedly will be a nightmare to audit, verify compliance and ultimately control and manage. One technology being eyed to untangle this spaghetti is a distributed consensus mechanism such as blockchain. Stop rolling your eyes—we’re not talking about Bitcoin or ICOs. We are describing the need for a centralized, permissioned, secure, distributed ledger to record, track, authenticate and audit the torrent of pipeline transactions. Blockchain seems to fit the bill nicely and also has the added benefit of enabling digital contracts at each pipeline stage that can be verified at runtime to trigger artifact promotion and also audited long after the fact to ensure that all rules were followed.
DevOps has already become the standard deployment paradigm in many enterprise IT shops and its adoption will continue to become more pervasive. But this growth requires DevOps products and processes to become smarter, more secure and more controllable as the release pipeline incorporates a wider set of deployment artifacts beyond just applications. This is a daunting challenge to be sure, but there are exciting new products that will emerge to help write the next chapters in the DevOps story.
VMware, at VMworld 2018 this week, advanced an ambitious campaign to automate IT management by announcing a deal to acquire CloudHealth Technologies and then previewing a Project Magna initiative that employs machine learning algorithms and other forms of artificial intelligence (AI) to create what VMware describes as self-driven data centers.
VMware CEO Pat Gelsinger told conference attendees that, as part of achieving that goal in the era of the cloud, every aspect of IT now needs to be automated.
“The rule of the cloud is to ruthlessly automate everything,” he said.
Project Magna represents an effort to centralize AI research and development to enable IT organizations to reallocate resources to building applications by eliminating many of the manual tasks that conspire to make IT inflexible.
At the same time, increased levels of automation will serve to reduce the cost of labor associated with managing IT, which remains the single highest cost within most IT budgets. Without increased levels of automation, those costs inevitably will rise as IT organizations embrace not only public clouds but also emerging new platforms such as Kubernetes.
Gelsinger this week made it clear VMware views being able to automate the management of Kubernetes as an opportunity to expand the reach and scope of its management offerings.
The addition of CloudHealth Technologies to the VMware portfolio adds a cloud operations platform that spans Amazon Web Services (AWS), Microsoft Azure and Google Cloud Platform (GCP). Once the acquisition is formally closed, VMware has pledged to integrate its cloud automation services, Secure State and Wavefront cloud services with the CloudHealth platform. The CloudHealth platform will be rebranded VMware Cloud Operations. Those moves come on the heels of a significant expansion of the portfolio of DevOps tools the company also announced this week.
VMware also revealed its is working on the Dimension project with Dell EMC and Lenovo, which will create a hyperconverged architecture optimized for deployment at the network edge. Project Dimension will make use of VMware NSX network virtualization software and the software-defined wide area networking (SD-WAN) platform gained by the company’s acquisition of VeloCloud to connect servers running at the network edge. That initiative complements the existing Pulse project for managing internet of things (IoT) devices, for which VMware announced REST application programming interface (API) support and which will be available as an on-premises or a software-as-a-service (SaaS) application.
VMware this week also unveiled updates to updates to its vRealize automation portfolio that enable organizations to rightsize workflows and automate host-based placement of those workloads. The core vRealize Automation platform also sports a new user interface and integration with open source Ansible Tower automation software and integration with Pivotal Container Service (PKS), an instance of Kubernetes. Version 7.5 of vRealize Automation 7.5 also features enhanced AWS and Azure support.
It’s unclear to what degree increased automation will transform how IT is managed within the context of modern DevOps processes. But the one thing that is clear is the economics of IT are being altered. Via automation VMware is sharply reducing the need for IT specialists in favor of “super” administrators capable of managing compute, storage, networking and security at unprecedented levels of scale in a much more integrated fashion. How much of that will be accomplished by AI is still anyone’s guess. But it’s already clear the job of the average IT administrator will soon be transformed utterly.
There’s been a lot of media attention in recent years about how artificial intelligence (AI) and machine learning (ML) are going to change the world—how they’re going to create new and interesting applications in fields as diverse as education, law, health care and transportation. This may happen. But if I had to bet on a use case where AI and ML will create a tangible, lasting impact, I’m putting my chips on DevOps.
DevOps is all about automation of tasks. Its focus is on automating and monitoring every step of the software delivery process, ensuring that work gets done quickly and frequently. While it doesn’t eliminate human tasks—far from it—it does encourage enterprises to set up repeatable processes that promote efficiency and reduce variability.
AI and ML are perfect fits for a DevOps culture. They can process vast amounts of information and help perform menial tasks, freeing the IT staff to do more targeted work. They can learn patterns, anticipate problems and suggest solutions. If DevOps’ goal is to unify development and operations, AI and ML can smooth out some of the tensions that have divided the two disciplines in the past.
Here are six ways AI and ML can and will change DevOps for the better.
Promoting Feedback on Performance
One of the key tenets of DevOps is the use of continuous feedback loops at every stage of the process. This includes using monitoring tools to provide feedback on the operational performance of running applications. This is one area today where ML is impacting DevOps already. Monitoring platforms gather massive amounts of data in the form of performance metrics, log files and other types. Advanced monitoring platforms are applying machine learning to these datasets to proactively identify problems very early and make recommendations. These recommendations go to the DevOps team members so that they can ensure that the application service remains viable. Machine learning is enhancing the continuous feedback loops that are critical to DevOps.
Enabling Communication
Communication and feedback is always one of the biggest challenges when organizations move to a DevOps methodology. Human interaction is vital, but with so much information flowing through the system, teams need to set up a wider variety of channels to set and revise workflows on the fly. Using automation technology, chatbots and other systems initiated by AI, these communications channels can become more streamlined and more proactive.
Correlate Data Across Platforms and Tools
To operate efficiently, DevOps teams need to simplify tasks. This is getting more difficult as environments get more complex. Start with monitoring tools: Teams tend to use multiple tools that monitor an application’s health and performance in different ways. Machine learning applications can absorb these data streams and find correlations, giving the team a more holistic view of the application’s overall health.
Manage a Flurry of Alerts
Since DevOps encourages teams to “fail but fail fast,” it’s critical to have an alert system that spots a flaw quickly. This tends to create scenarios where alerts are coming fast and furious, all labeled with the same severity, making it difficult for teams to react. Machine learning applications can help teams prioritize their responses based on factors such as past behavior, the magnitude of the current alert and the source that specific alerts are coming from. Humans can set up rules, but machines can help manage these types of situations when too much data overwhelms the system.
Evaluating Past Performance
AI/ML also has the potential to help developers during the application creation process. By examining the success of past applications in terms of build/compile success, successful testing completion and operational performance, machine learning algorithms could make recommendations to developers proactively based on the code they are writing or the application that they are building. The AI engine could direct the developer in how to build the most efficient and highest-quality application.
Software Testing
In the future, we could see AI/ML applied to other stages of the software development life cycle to provide enhancements to a DevOps methodology or approach. One area where this may happen could be in the area of software testing. Unit tests, regression tests, functional tests and user acceptance tests all produce large amounts of data in the form of test results. Applying AI or machine learning algorithms to these test results could identify patterns of poor coding practices that result in too many errors caught by the tests. This information could then inform the development teams so that they can become more efficient in the future.
Similarly, leveraging historical data, AI/ML could be used to fine-tune deployment strategies as applications are moved from Dev to Test to Production environments.
According to McKinsey’s Global Institute Report of 2018, artificial intelligence (AI) has the potential to create an annual value of $3.5 billion to $5.8 billion across different industry sectors. Today, AI in finance and IT alone accounts for about $100 billion; hence, it is becoming quite the game changer in the IT world.
With the onset of cloud adoption, the world of IT DevOps has changed dramatically. The focus of ITOps is changing to an integrated, service-centric approach that maximizes business services availability. AI can help ITOps in early detection of outages, potential root cause prediction, finding systems and nodes that are susceptible to outages, average resolution time and more. This article highlights a few use cases where AI can be integrated with ITOps, simplifying day-to-day operations and making remediation more robust.
Predictive Analytics of Outages
False positives can cause threat alert fatigue for ITOps teams. One survey indicates that about 52 percent of security alerts are generally false positives. This puts a lot of pressure on the teams, as they have to review each of these alerts manually. In such a scenario, deep neural networks can predict whether an alert will result in outages.
Alerts Layers Yes/No
Feed Forward back propagation with two hidden layers should yield good results in terms of predicting outages as illustrated above. All alert types within a stipulated time can act as inputs and outages would be the output. Historical data should be used to train the model. Every enterprise has its own fault line and weakness, and it is only through historical data that latent features are surfaced; hence, every enterprise should build its own customized model, as a “one size fits all” model has a higher likelihood of not delivering expected outcomes.
The alternate method is a logistic regression where all “alert types” are input variables and “binary outages” would be the output.
Logistic regression measures the relationship between the categorical dependent variables and one or more independent variables by estimating probabilities using a logistic function, which is the cumulative logistic distribution. Thus, it treats the same set of problems as probit regression using similar techniques, with the latter using a cumulative normal distribution curve instead.
Root Cause Classification and Prediction
This is a two-step process. In the first step, root cause classification is done based on keyword search. From free-flow root cause analysis fields, natural language processing (NLP) is used to extract key values and classify into predefined root causes. This can be either supervised or unsupervised.
In the second step, random forest for multi-class neural networks can be used to predict root causes while other attributes act as input. Based on the data volume and the datatype, one can choose the right classification model. In general, random forest has better accuracy, but it needs structured data and right labeling and it is less fault-tolerant to data quality. While a multi-class neural network will need a large volume of data to train, it is more fault-tolerant but slightly less accurate.
Prediction of Average Time to Close a Ticket
A simple weighted average formula can be used to predict time taken for ticket resolution:
Avg time (t) = (a1.T1 + a2.T2+ a3.T3 )/(count of T1+T2+T3)
Where T1 are ticket types.
Other attributes can be used to segment the ticket into right cohorts to make it more predictable. This helps in better resource planning and utilization. Weightage of features can be done heuristically or empirically.
Unusual Load on System
Simple anomaly detection algorithms can inform whether the system is going through a normal load or it has high variance. A high variance/deviation from average on time series can inform the unusual activities or resources that are not freeing up. However, the algorithm should take care of seasonality, as a system load is a function of time and season.
Given the above scenarios, it is obvious that AI has a tremendous opportunity to serve IT operations. It can be used for several ITOps including prediction, event correlation, detection of unusual loads on system (e.g. cyberattack) and remediation based on root cause analysis.
Artificial intelligence (AI) and machine learning (ML) can help the humans in DevOps break free from focusing on simple activities. One aspect of DevOps is automating routine and repeatable actions, and AI and ML can perform these activities with enhanced efficiency to improve the performance of teams and business. There are algorithms that can perform many operations and procedures, allowing those in DevOps to execute their part effectively. This article discusses how DevOps engineers can use AI and ML to their benefit.
Businesses are under a lot of pressure to meet customers’ ever-changing demands, and many embrace DevOps to improve their performance to some extent. However, it can be difficult for many companies to use AI and ML because of the complexity involved. To recognize any benefit with AI and DevOps, a creative mindset may be required.
The adoption curve of AI/ML may be relatively slow. Only 27 percent of CIOs surveyed by ServiceNow for its report, “The Global Point of View,” have hired employed who have skills in machine learning. But the fact is, DevOps experts may have a lot to gain by adopting even the most basic features of AI and ML. The same survey found that around 85 percent of C-level executives believe AI/ML can offer substantial value in terms of accuracy and rapidity of decision-making, which will lead to improved profitability for the company.
Tracking and organization in a DevOps environment requires effort because of the complexity involved in the distributed application, which traditionally made things difficult for the team to manage and resolve customer issues. Before the evolution of AI and ML, DevOps team members could spend hundreds of hours and a large amount of resources to identify one point within an exabyte of information. To solve such problems, the future of DevOps is AI-driven, helping to manage the immense capacity of data and computation in day-to-day operations. AI has the potential to become the primary tool for assessing, computing and decision-making procedures in DevOps.
AI’s Influence on DevOps
AI can change how DevOps teams develop, deliver, deploy and organize applications to improve the performance and perform the business operations of DevOps. There are three common ways through which AI may influence DevOps:
Enhanced Data Accessibility
The lack of unregulated accessibility to data is a critical concern for DevOps teams, which AI can address by releasing data from its formal storage—necessary for big data implementations. AI can collect data from multiple sources and prepare it for reliable and robust evaluation.
Greater Implementation Efficacy
AI contributes to self-governed systems, which allows teams to transition from a rules-based human management system. This helps address the complexity of assessing human agents to improve efficacy.
Effective Resources Use
AI gives much required competence to automate routine and repeatable tasks, which minimizes the complexity of managing resources to some extent.
How Can Companies Apply AI and ML to Optimize DevOps?
Organizations can apply AI and ML to greatly optimize their DevOps environment. For one, AI can help in managing complex data pipelines and create models that can feed data into app the app development process. By 2020, it’s expected AI and ML will take the lead in digital transformation, overtaking IoT.
However, implementing AI and ML for DevOps also presents a number of challenges for organizations of all sizes. To benefit from AI and ML technologies, a customized DevOps stack is required.
Open source projects such as the Fabric for Deep Learning (FfDL) and Model Asset eXchange (MAX) can lower the barrier of entry for companies, helping to implement machine learning and making the DevOps process more efficient.
Application of AI and ML can result in true ROI for a company by optimizing DevOps operations, making IT operations more responsive. They can improve efficiency as well as productivity of the team and play an important role in filling the gap between humans and big data.
Conclusion
A company that wants to automate the DevOps have to decide whether to buy or build a custom AI/ML layer. However, the first step is to establish a strong DevOps infrastructure. Once the foundation is created, AI/ML can be applied for increased efficiency. AI/ML can help DevOps teams focus on creativity and innovation by eliminating inefficiencies across the operational life cycle, enabling teams to manage the amount, speed and variability of data. This, in turn, can result in automated enhancement and an increase in DevOps team’s efficiency.
Reliance on machine learning algorithms and other forms of artificial intelligence (AI) to not just run application tests at scale but actually construct them may be about to increase.
The latest release of a Digital Automation Intelligence (DAI) Suite from Eggplant enhances the AI capabilities baked into the testing tool to include the capture and analysis of runtime performance data in a way that enables a testing process to “learn” where potential bottlenecks are hidden. Eggplant CTO Antony Edwards said DAI Suite will then suggest a series of tests designed to make it easier for developers to identify the root cause of the problem.
The goal is to reduce reliance on humans to construct formulaic tests, which Edwards said often limits the rate at which applications can be deployed and updated.
Eggplant’s DAI Suite
Other new capabilities being added to DAI Suite include enhancements to Eggplant Functional, which now takes advantage of AI to discover the elements of a graphical user interface (GUI) to automatically populate the automation logic in an Eggplant AI application model. Eggplant Performance, a load and performance testing product, has been integrated with JMeter, an open source performance testing framework, to enable developers and testers to reuse JMeter assets as well as creating load test scripts. Additionally, new real-time key performance indicators (KPIs) can be employed to shorten the feedback loop between testing and analysis.
Both Eggplant Manager and Eggplant Automation Cloud are now 10 times faster and provide higher levels of testing scalability, according to the company.
Edwards noted that, in general, application testing is too formulaic. The end goal should be on identifying the best way to delight an end user rather than merely trying to check a compliance box. That requires being able to test more application use cases at scale without slowing down the overall application development life cycle. In fact, he said the biggest hindrance to DevOps today is a reliance on manual testing processes that are driving a $32 billion testing services sector.
It’s now only a matter of time before AI is applied more broadly to application testing. As the number of applications being developed increases, the more challenging testing becomes. The longer it takes to conduct those tests the more tempted developers are to short-shrift the testing process altogether, believing whatever issues that might arise can be addressed quickly using agile development methodologies. However, not only are there more dependencies than ever between applications, but also end users are less forgiving. Many of them will not, for example, ever give a mobile application that initially disappoints them a second look. Compounding that pressure is increased reliance on digital business processes, where the quality of the underlying application experience directly correlates to the perception of the value of the corporate brand.
Developers may not always have the highest regard for application testing. But as IT continues to evolve, chances are high there soon will be many more developers perfectly happy to take advantage of all the AI help they can get.
Saying that new jobs will always be created to replace those that have been eliminated by automation is a mechanism for avoiding a real problem that’s lurking on the horizon.
In 2004, at the height of its popularity, Blockbuster Video had more than 9,000 stores and employed more than 80,000 people worldwide. About 50,000 of the total workforce was employed in the United States. Going down to Blockbuster to pick up a movie was a Friday night ritual for many, this writer included.
Blockbuster and other video rental stores, along with the advent of affordable Big Screen TVs, were significant contributing factors in the growth of the home theater. People now had an alternative to $10 movie tickets and battling weekend traffic to get to the cineplex. An entire family could view a recent movie in the comfort of their own home for about $5. The only thing that was needed to complete the movie going experience was the popcorn, which frugal families bought at Costco by the caseload. Those who were less thrifty could buy a box of microwave popcorn at the checkout counter of the video store. The $25 cost of a Friday night date was reduced to roughly $6.
Blockbusters kept people on the couches and out of the movie theaters. Television had already put a damper on movie-going. In 1930, 65 percent of the population went to the movies once a week. By 1964, the number dropped to 10 percent. Despite the introduction of 3D projection technology and movie theaters offering added features such as in-seat dining, movie attendance is now at its lowest point since 1992.
Blockbuster was a transformational business on the entertainment landscape. Today, the company is gone. Streaming technology combined with an “all you can eat” subscription model killed it. (See Figure 1.)
Figure 1: As Blockbuster stock declined, Netflix’s grew.
Just as Blockbuster did away with the need to go to a movie house to see a film, Netflix, Amazon and other video streaming services did away with the need to leave the house to get the evening’s entertainment. America stayed home and enjoyed movies on the internet for a fixed monthly fee. Blockbuster went out of business.
Netflix employs around 5,500 people with gross yearly sales of $11.69 billion. Blockbuster’s gross sales in 2004 were $6 billlion. It’s astounding that Netflix does almost twice as much business as Blockbuster did at its height, with less than 10 percent of the Blockbuster workforce. Such is the benefit of automation.
Whereas Blockbuster needed to staff a store in every neighborhood to do business, Netflix and other streaming services work from a central, virtual location without the added expense of redundant staffing. Blockbuster had 9,000 cash registers to man. Netflix has one, which is run by some sort of AI.
So, where did all the out of work clerks and store managers go? Hopefully, they went on to other jobs, maybe ones that were far better than the day-in, day-out tasks that go with processing video cassette rentals. After all, that’s the conventional wisdom: There will always be new jobs to take the place of the ones eliminated by automation. But, let’s face a cruel truth: It takes a lot more knowledge and skill to work on video delivery at Netflix than it does to check out customer purchases at Blockbuster. And Netflix requires fewer people to make the business go. Would a Blockbuster employee in Sioux City, Iowa, have the skills and mobility necessary to go on to to work at Netflix in Los Gatos, California? Seems pretty far-fetched.
Right now, things are pretty rosy in terms of the economy. So the conventional wisdom prevails: As automation eliminates one job, another new type of job takes its place.
Yet, there is another way to look at it. There are good arguments to be made:
As automation eliminates jobs in the modern world, new jobs for humans will NOT be created at levels required to keep a workforce fully employed. Rather, AI will do more and require less.
The lead up to large-scale unemployment will not be slow. It will happen fast, over a period of a year or two. And, it very well might happen within a decade rather than decades in the future.
Productivity due to automation continues to accelerate. In the early 1980s, it took 10.4 man-hours of labor to make a ton of steel. By 2014, the number was down to 1.9 man-hours per ton. The increase in productivity is due to increased levels of automation (think: robots). Given the choice between buying a machine and hiring a worker, companies are more likely to choose the machine. Thus, it’s entirely possible to imagine that the day when a few humans will be supervising the work of many robots making steel is close. Not too soon after that, a single human will be supervising an entire robotic workforce. Then, finally, as AI continues to improve and further dominate the production landscape, no humans will be required. The robots will supervise themselves.
How far off in the future is this time? Dunno. But I can tell you that robots are now assisting with surgery on humans. How long will it be until one surgeon can use automation to do the work of 10? Then, how long will it be until AI can do surgery unsupervised, five years? Ten years? Only 10 years ago driverless cars were still on the test tracks; today they are roaming the streets of Mountain View, California. An autonomous robot with a scalpel is no longer the stuff of science fiction. Once AI can do surgery unassisted, where will all the displaced surgeons find work? Maybe a few will go into research. What of the others?
At one time there were 200,000 horses in the City of New York. They provided the labor for a good part of the city’s transportation system. Today there are fewer than 1,000 horses. Cars, trucks, planes and trains have taken over. There’s no benefit to employing a horse when a car can do the work better. As General AI continues to develop and improve its ability to connect the dots as only a human can, how long will it be until the labor of most humans is not needed? What will it be like when there is no benefit to employing large parts of the human population because a machine can do the work better? Yes, the ramifications are scary. But, the possibility needs to be considered. The sad fact is that few people are considering such a scenario, let alone coming up with ways to address the impact on the horizon. Saying there will always be employment for humans is an avoidance mechanism, particularly given the fact that modern machines are going beyond purely mechanical labor and doing more thinking work.
In 1990, people thought there would always be movie rental until there wasn’t. In 2000, people thought there would always be a Virgin Megastore until there wasn’t. Today, Sears keeps closing stores. It will soon be gone from the face of the earth. Yes, Amazon is opening brick and mortar stores. This might be good news. But the fact that these stores are highly automated and have few employees cancels out the good feelings of the moment.
These days when a company comes to an end, it’s not a drawn-out process. It happens fast, over a period of a year. Many of the people left behind are confronted with low-wage, “show up and do it” jobs. The more educated have access to a pool of better paying work that comes with an excruciatingly steep learning curve. Whether the job is low-wage retail work or high-paying tech work, each worker is—or will be—competing against AI-powered automation that can—or will—do the job better.
It’s not a pretty picture. But it is one that can avoided if we decide to think about alternatives and put those ideas into action. Sadly, there are more people thinking about how to get to Mars than solving the problem of unemployment on a massive scale due to the proliferation of industrial automation powered by general AI. The problem is that we still seem to be stuck at, “There will always be jobs to replace the ones eliminated by automation.” To say otherwise is deemed Cassandra talk. Maybe so. But to not give the problem consideration is beyond folly. It’s a danger that will exact a heavy price from everyone who has to work to earn a living.
Figure 1: As Blockbuster stock declined, Netflix’s grew.