White Paper

Scaling Modzy

You’ve just spent a lot of time and effort building an amazing model and it works perfectly on your machine. Now comes the next challenge – how do you make it useful for your entire organization? How do you feed a mind-bogglingly large amount of data through it? How do you track its performance once you figure out answers to the first two questions?

What You Need to Know

Putting software into production is hard. Not only are there the technical challenges of making sure that the right code runs on the right machine at the right time, but there are the added challenges that stem from ensuring compliance with your organization’s requirements.

Scalability is about more than just running your application many times behind a load balancer. As you increase the scale of your application deployment, you also increase the need for infrastructure to monitor, alert, secure, and report on the health of your application. Likewise, in the world of container orchestration, scalability is also about usage efficiency. Cloud providers make near-infinite scale possible with the push of a button, but infrastructure budgets and on-premise infrastructure have not evolved in the same manner. Highly scalable software must also be highly efficient to ensure effective use of existing physical resources and prevent unnecessarily ballooning cloud infrastructure costs.

The next challenge stems from data. One of the greatest benefits of artificial intelligence (AI) and machine learning software is its ability to parse and analyze volumes of data at a speed and scale far faster and great than that of human cognition. Data can come from a variety of different locations in a variety of formats, and can also have different access and security requirements.

Modzy Differentiation

The first way that Modzy helps you achieve scale is by not requiring your data scientists and model writers to know how to do it. Our model interface abstracts away all the complexities of running a model at scale with a simple interface that can be leveraged by anyone with a little bit of coding experience. Model authors simply define model hardware requirements, and implement a simple 3-route web server. Inputs and outputs are handled by reading and writing from the filesystem.

Modzy handles the intricate complexities of running software at scale automatically. We take the hardware requirements for each model and ensure that models are run on machines that can meet those requirements. Modzy also fetches the data from where it lives, and presents it to the model through its filesystem so that model authors never need to write data connection adapters, or worry about data access security, logging, or auditing.

Finally, Modzy takes advantage of the full power and capability of Kubernetes to support horizontal scaling, as well as efficient resource utilization. Because we’re built on Kubernetes, there is a veritable wealth of supporting software to actively monitor, secure, alert, and report on real-time operation.

What This Means for You

Modzy helps anyone author amazing models that can run at enterprise and cloud scale, without requiring IT, security, networking, or policy expertise. Modzy supports both the near-infinite scalability of the cloud and the efficient utilization needed for on-premise installations. Because we’re built on ubiquitous, community-supported software, Modzy can tie into all your existing IT infrastructure to monitoring and maintenance in accordance with your organization’s requirements.

Get Details on Quick Start Program