Alluxio launches orchestration layer optimized for AI, ML

New capabilities

In the 11 months since OpenAI released ChatGPT, which represented a leap in generative AI and large language model capabilities, many data management and analytics vendors have made generative AI the primary focus of their product development.

But just as vendors are striving to develop generative AI tools such as text-to-code translation and anomaly detection, organizations are developing their own generative AI models to improve productivity.

Those models as well as traditional AI and machine learning models benefit from easy access to relevant data. They also require significant compute power and storage capacity that isn't always readily available.

Many data infrastructures, however, don't enable users to easily find and retrieve the most pertinent data for a given AI or machine learning model. They don't have the GPU resources needed to provide enough compute power and storage capacity required by complex AI and machine learning models.

Alluxio aims to provide that needed infrastructure with Enterprise AI and is a significant new platform for the vendor's customers, according to Kevin Petrie, an analyst at Eckerson Group.

The platform, which sits on top of the data lakes organizations use to store their disparate types of data, is specifically designed to enable high-performance model training and deployment, data access across regions and clouds, unlimited scale, and the ability to work with an organization's existing data storage rather than requiring specialized data storage.

"These enhancements improve performance and scalability," Petrie said. "This is becoming increasingly important for companies that need to train, fine tune and operationalize large language models or other advanced AI models."

Enterprise AI includes the following:

• A decentralized system architecture that is designed for infinite scale.
• Access to data through a single portal that connects the decentralized system architecture and enables collaboration across departments.
• New APIs for model training and servicing that enable concurrency to deliver increased performance.
• A distributed caching model that enables AI engines to read and write data through Alluxio's cache rather than those in data lakes.

The result is a data access layer that enables customers to simplify and accelerate AI and machine learning workloads using multiple compute engines, such as Presto and PyTorch, Petrie noted.

Combining that with reading and writing to data in multiple locations, a security framework, open APIs, and other capabilities makes Enterprise AI a unique orchestration layer, he continued.

"I'm not aware of other companies that have capabilities like Alluxio," Petrie said. "The primary alternatives for companies are to build and implement their own tools or take advantage of platform-specific tools that optimize workloads in specific locations."

Madan, meanwhile, noted that Alluxio customers were able to train AI and machine learning models with Enterprise Data. But the platform's approach could not meet the scale required by AI and machine learning models.

In particular, the architecture and interface weren't suitable for AI and machine learning. "Some people could do it," Madan said, "but not effectively."

Therefore, following requests from customers to better enable AI and machine learning model development, Alluxio initially attempted to enhance Enterprise Data before deciding to build an entirely new platform optimized specifically for AI and machine learning, he continued.

These enhancements improve performance and scalability. This is becoming increasingly important for companies that need to train, fine tune and operationalize large language models or other advanced AI models.

"We were incubating some of these functionalities," Madan said. "But because of our open source offering, people were coming in and asking us to do [more]."

Alluxio is targeting a different set of users with Enterprise AI than it does with Enterprise Data, he added.

The vendor's previous tools were meant for data administrators and others tasked with overseeing their organization's data operations. In addition, Enterprise Data is geared mainly toward large enterprises like Meta and Uber, which Madan said are both Alluxio customers.

Enterprise AI instead targets those who oversee their organization's AI operations. It can be used by small and mid-sized organizations as well as large ones.

"It's a new segment for us," Madan said. "They tend to be a lot smaller than the large enterprises that have very large data lakes. The new persona we're targeting is folks that are only starting to scale out from training models on a single server to using distributed training."

Looking ahead

With Enterprise AI now generally available, one of Alluxio's aims going forward is to provide users with insights about their data, according to Madan.

The vendor's data orchestration platforms sit in the middle of the data stack between data storage and data analysis. Given that position, Alluxio's tools are able to work with data both where it rests and where it's operationalized, which enables it to derive insights from throughout the data lifecycle.

Petrie, meanwhile, suggested that Alluxio should enhance its platform with generative AI capabilities that make data workers more productive.

Vendors such as Boomi, Dremio and MongoDB are among the many data management and analytics specialists that have such features in the works with some already generally available.

"I'd be interested to see future enhancements that further leverage generative AI to improve ease of use, such as providing a conversational interface to users to guide and assist configurations," Petrie said.

Eric Avidon is a senior news writer for TechTarget Editorial and a journalist with more than 25 years of experience. He covers analytics and data management.