At the LEAP East conference, openEuler showcased its SuperPoD OS, designed to improve performance for AI workloads running on clustered computing systems. The system, built around the UnifiedBus architecture and UB Service Core software, aims to coordinate resources across multiple machines, allowing them to function as a single, powerful unit.
openEuler’s SuperPoD OS extends operating-system support for UnifiedBus-based SuperPoD computing systems. These SuperPoDs are clusters of powerful computers intended to operate as a unified machine. Coordinating these machines for AI tasks, sharing memory and data, can be complex. UB Service Core, part of the openEuler ecosystem, provides the necessary coordination through five cluster-level system services: Engine, Memory, Communication, I/O, and Virtualization.
The Engine service pools memory and DPU resources, dynamically scheduling them and providing automatic failover. The Memory component enables machines to share and pool memory using unified-memory programming, presenting distributed memory as a single addressable space. Communication provides high-performance networking compatible with existing Socket and Verbs interfaces. The I/O service manages global data caching for AI inference workloads, and the Virt component handles virtualization pooling, live migration, disaster recovery, and communication efficiency between VMs and containers.
According to openEuler, UB Service Core can improve application performance by 30-50% by leveraging the peer-to-peer UnifiedBus interconnect architecture. The entire SuperPod system and UB Service Core are fully open source under the openEuler umbrella. In addition to infrastructure, openEuler is also integrating AI directly into its operating system. The DevStation desktop variant of openEuler 24.03 LTS SP4 includes openEuler Intelligence, a desktop AI assistant utilizing an agent runtime and MCP-based tools for development and system operations.
Read the original coverage
💬 Comments
📜 Comment Policy