Oracle announced new networking capabilities for its Oracle Cloud Infrastructure (OCI) service on Oct. 27, adding front-end Remote Direct Memory Access (RDMA), broader NVMe block-storage options and network path recovery features for distributed AI workloads. The additions extend Oracle Acceleron, Oracle’s networking and infrastructure technology suite.
Oracle said the new High-Performance Fabric brings RDMA connectivity to OCI front-end compute for inference, storage and high-performance computing. The company also expanded NVMe block-storage attachments across Oracle Acceleron bare-metal and virtual-machine shapes. Oracle says the storage changes can improve performance by up to 3x, but did not specify the comparison baseline or measurement method.
For distributed AI communication, Oracle’s Collective Communications Library spreads traffic across OCI’s backend Multiplanar Network and steers it toward healthier paths. Oracle says the system can help workloads recover when links, ports or network interface cards fail. A separate front-end networking capability adds path selection, failover and congestion control for RDMA traffic.
Oracle engineers say synchronized AI jobs can be slowed by a single congested or unreliable network path, potentially leaving GPU capacity underused. Oracle’s engineering blog describes RDMA as allowing machines to access one another’s memory with little CPU work, a design intended to reduce delays and sustain throughput.
Oracle did not provide feature availability dates, supported regions or pricing. Its announcement says product timing and pricing may change and does not commit the company to deliver the described features.
