Data Centric Computing

Project Lead

Yang Seok Ki

Data-Centric Computing has emerged as a critical paradigm shift in response to the limitations of traditional CPU-centric computing models. As CPU performance improvements based on Dennard scaling and Moore's Law have reached their limits, the challenge of handling large amounts of data efficiently has become paramount. This workstream focuses on exploring and developing Data-Centric Computing technologies that prioritize optimal computation placement based on data location and computational complexity rather than simply moving all data to the CPU for processing.

The shift towards Data-Centric Computing is driven by the need for power-efficient data handling and the limitations of bandwidth in traditional models. This approach builds upon the success of domain-specific computing architectures, such as graphics and network cards, and has expanded to include various accelerators. In environments where power efficiency is crucial, such as hyperscaler data centers, ASICs and FPGAs are increasingly used to offload and accelerate OS, security, and data processing tasks.

Scope

The scope of this workstream encompasses the horizontal integration of various Data-Centric Computing technologies. While not focusing on defining individual technologies, this workstream aims to create a cohesive framework that allows different data-centric approaches to work together seamlessly. The technologies under consideration include:

1. Computational Storage: Technologies that bring computation closer to data storage locations.
2. In-Memory Processing: Techniques for performing computations directly within memory arrays.
3. Near-Memory Processing: Architectures that position computational units in close proximity to memory.
4. Data Processing Units (DPUs): Specialized network processors designed for data-centric tasks.

The primary focus of this workstream is on developing methods for horizontal integration across these technologies, enabling their combined use in a more natural and efficient manner. To achieve this, the workstream addresses three key areas:

1. Common Computing Interface: Developing a standardized interface that allows heterogeneous data-centric technologies to communicate and interact effectively.
2. Resource Discovery and Management: Creating systems and protocols for identifying, allocating, and managing diverse data-centric computing resources within a unified framework.
3. Sustainable Scaled Computing: Ensuring that the integrated data-centric computing solutions can scale efficiently while maintaining sustainability in terms of power consumption and resource utilization.

By concentrating on these aspects, the workstream aims to foster a more integrated and flexible data-centric computing ecosystem, capable of leveraging the strengths of various technologies in combination to address complex data processing challenges.

OCP FTI Data Centric Computing Calendar

The calendar displayed here is updated nightly from the project's Groups.io Calendar