Google hosted over 2,000 people in San Francisco this week at the much anticipated Google NEXT conference. One of Google’s goals for this conference was to show how serious the company is in winning over enterprise customers. It spent a large amount of time in the Day Two keynote talking about the enormous amount of security that goes into building and running its world-class data centers. It also impressed the audience with its unmatched capabilities in high-performance computing and scalability. The message was clear that for enterprises the Google Cloud Platform was the place to go for both price and performance.
Schmidt went on to make a few more predictions:
- The platform is not the end, it is the bottom. Machine learning is on top and will be the technology that will drive the transformation.
- NoOps will become mainstream.
- Serverless architectures will be the next wave of computing.
Machine Learning
Google has been blazing the path for big data for almost two decades. Now it wants to shift the conversation from big data analytics to deep learning. Google believes machine learning is the next layer of programming. Its goal is to make the process of data ingestion, storage, and training machine models as simple as calling an API. This will allow developers to focus on creating incredible new applications without having to understand complex concepts like neural networking.
We’ve known for years that most big data initiatives fail in the data ingestion phase. Google has made this process very easy with APIs like Pub/Sub, DataFlow, and others. We also see data scientists complaining that they spend up to 80% of their time preparing the data and training the models before they can even begin to extract any value out of the current machine learning technologies. In fact, some data scientists sarcastically call themselves “data janitors” because they spend more time preparing data than they do analyzing it.
Google showed us its speech, sound, and image recognition services. The Google Cloud Vision API can analyze an image and categorize its contents into thousands of categories. For example, the API can take an image of a person and detect race, gender, and mood. It can also look at the background and possibly determine the location of the person and relevant information about the location. It can also read text within the image itself and return it to the user as metadata.
The applications of the Vision API are endless. Imagine a healthcare company working with images generated from X-rays and MRIs. The Vision API can return various metadata about what it sees within each image. Then machine learning models can be trained to make predictions based on the image metadata. When applied to industries like healthcare, image recognition and machine learning can could lead to new methods of early detection of diseases like cancer. My guess is that these technologies will be used to solve problems and predict outcomes for things that we haven’t even thought about yet.
NoOps
When Schmidt predicted that NoOps will become mainstream, I tweeted it. My Twitter stream immediately blew up with an all-out war on NoOps versus DevOps. The point that Schmidt was really trying to drive home is that in the future, developers will leverage more higher level services, like Google’s machine learning services, so they won’t have to care about designing, building, scripting, and managing the underlying infrastructure and software that supports the API. Unfortunately, the word “NoOps” is often perceived as we no longer need operations, or we no longer need to do DevOps.
Schmidt made the point that today’s programming model is based on 20-year-old processes that force developers to think about infrastructure. The new model should allow developers to focus on business requirements, while the cloud provider handles the infrastructure and the scalability. The more the underlying technologies are abstracted, the less operational tasks are required, hence the NoOps term. Companies still need operations, but the focus should shift towards operating the applications and ensuring that the cloud platform is meeting its SLAs. That is a very different operating model than running data centers and middleware.
Serverless Architectures
Another reason Google thinks that NoOps will go mainstream is that future architectures will follow the serverless architecture pattern. This means that the platform will dynamically determine how much infrastructure is needed and then automatically provision and deprovision the infrastructure to support the application.
Think about how much time is saved when the upfront efforts of building and managing infrastructure are removed from the software development life cycle (SDLC). In today’s environment, large amounts of time and resources are invested in building out systems that can scale and failover. Google’s serverless architecture approaches abstracts that work and gives the developer the quickest path from ideation to production--running on top of the same technology and processes Google uses to run applications like YouTube, Gmail, and others.
Summary
Cloud computing has come a long way over the last several years. Innovation is increasing at rates never experienced before and will continue to do so. Cloud computing started as a way to abstract physical infrastructure and data centers. The next generation of cloud computing will focus on abstracting virtual infrastructure and the operational processes that go along with managing that infrastructure. Complex technologies like machine learning will be made simple, so data scientists can focus on the data, and developers can enable the data scientists without having to become experts at the underlying machine learning technologies.
As machine learning becomes easier to work with, companies will shift from performing analytics to deep learning. These technologies will undercover new information that can transform industries and business models. This is an area where Google has a clear competitive advantage over the other cloud vendors because the new services they’re bringing to market have been proven and used internally at Google for years. Now they are delivering these services in a way customers can easily consume.
