- Industry: Commercial, Information Technology
- Type: Artificial Intelligence and Machine Learning
- Location: Minneapolis, MN, Worldwide
- Technical Category: Software, Hardware, Cloud
- IT status: Implementation of VMware and Nvidia as core of machine-learning-as-a-service platform, while meeting aggressive product-launch timeline
Sterling helps cyber firm neurothink offer cutting-edge MLaaS
neurothink, a Minneapolis-based artificial-intelligence company that is coming out of development in early 2022, is getting advanced compute capabilities from next-generation VMware Cloud Foundation, Tanzu Kubernetes, and Nvidia. neurothink specializes in machine-learning-as-a-service (MLaaS) in a new and radically accessible way, by taking on the technical overhead for data-science-based clients who want to run complex analytical models with quick and direct access to powerful and costly hardware and software.
According to neurothink Senior Cloud Architect Akhilesh Miryala: “Without the implementation of VMware, neurothink would not have the exceptional accessibility it has, which differentiates the neurothink brand from other machine-learning technologies. neurothink is precisely ML made more available, more affordable, and more streamlined—with friendlier interfaces—all while being more secure than other brands’ technologies.”
The custom, uncluttered interface and automated, virtualized back-end give customers efficient and quick access directly through a browser window, with nothing to install. Clients can stay in their development environment as long as they want, for one fee, before they are ready to access enterprise-grade GPUs for model training. neurothink thus offers a ready solution for a range of researchers: from expert programmers who want simpler access to a command-line interface to build and run complex experiments, to newer data scientists more comfortable using notebooks.
In turn, this could accelerate information breakthroughs. “Researchers will tap more readily into the promise of machine learning, able to test their models and proofs-of-concept, exploring massive and extensive quantities of data, discovering and mining what they had not explicitly programmed to unearth,” Miryala added.
Sterling cloud-architects Billy Downing and Nathan Bennett helped neurothink specify, procure, and implement the VMware and Nvidia software and systems that are at the core of the machine-learning platform. “Without Sterling’s technical expertise in using VMware,” says Miryala, “We could not have met the aggressive timeline of our fall product demonstration or other startup goals.”
According to Downing, “Sterling assisted neurothink in creating the underlay networking components that would connect the disparate hardware components, and all within a centralized platform, via VMware Cloud Foundation with Tanzu.”
Fellow Sterling cloud architect Bennett continues, “We then were able to layer this centralized platform with software-defined networking, using VMware NSX-T to remove the complexity in the physical devices and bring them into a software-defined network. Any future changes to network infrastructure that would have formerly required tedious, time-consuming physical work, servicing, and hardware, will be realized through this software definition.”
Sterling provided neurothink with guidance in establishing best-practice design and implementation of VMware Cloud Foundation (VCF) as their private cloud infrastructure, Tanzu Kubernetes for application hosting, and Bitfusion to access Nvidia Graphics Processing Units (GPUs) over the network backbone.
Providing a convenient and secure-access method to a dense population of GPUs for machine learning has traditionally been met with several challenges — the first, the ability to automate provisioning of isolated workspaces without excessive overhead. But VCF coupled with Tanzu solved this issue by isolating workloads within Kubernetes pods and virtual machines that are provisioned at runtime and scaled based on load. Tanzu’s direct integration within vSphere allows neurothink to closely monitor resource-utilization and determine appropriate access methods for administration, without their having to adopt an entirely new skillset outside their existing VMware expertise.
A second common issue is efficiently utilizing GPUs to meet the specific workload demand, without over- or under-provisioning resources. Bitfusion provides the capability to virtualize GPUs and fractionalize allocation of compute power to precisely meet the application-specific demand; therefore, a customer requesting a quarter of a single GPU can receive just that amount— without tying up the entire resource for their workload. Bitfusion also abstracts the location of the workload itself away from the underlying hardware by enabling network-based access to GPUs that are hosted across several independent servers, imparting a resilient architecture where GPUs are not aggregated to a single host subject to failure.
This solution provided neurothink the capability to offer their customers dynamic access through an intuitive web interface to a shared pool of GPU resources that are efficiently utilized by provisioning only the amount of processing required (a fractional of a single GPU for development or multiple GPU cards aggregated for the most demanding model training). This is essential to the neurothink model, since GPU fractionalization significantly lowers the total cost of using enterprise-grade tools, thus making these tools more accessible to everyone.
Tanzu Kubernetes provides an ideal application platform to enable automated scalability, customer isolation, and ease of management — all directly integrated within the VMware Cloud Foundation portfolio, including vSAN for stateful application-data retention, and NSX-T for software-defined networking security, visibility, and orchestration.
Finally, according to Miryala, “It’s not enough to do the right thing at the right time. You also have to do it the right way, and Sterling with their expertise helped us accomplish our goal of developing a first-of-its-kind platform the right way and on time.”