Oracle bolsters cloud plumbing for distributed AI training
Article excerpt
Highlighted: the sentence this signal was extracted from
Launches OCCL to improve network efficiency for next-gen workloads Oracle released a network plugin dedicated to distributed AI training. Dubbed the Oracle Collective Communications Library (OCCL), the tool integrates with similar libraries from Nvidia and AMD like NCCL and RCCL, respectively, through their standard network-plugin interfaces. The tool supports remote direct memory access (RDMA) over converged Ethernet version 2 (RoCEv2) and InfiniBand, and monitors queue-pair transfers and dynamically distributes traffic across network paths with less congestion. OCCL also automatically reroutes traffic around supported link or network interface card (NIC) failures and restores connectivity when hardware recovers. The software supports single-planar and multiplanar RDMA networks in a vendor-agnostic fashion. Oracle said this provides a software-defined way to better support collective-communication resilience and traffic distribution while maintaining flexibility in the underlying network hardware. According to internal benchmarks using Nvidia GB300 systems with 16 to 128 nodes and four GPUs per node, OCCL increased average NCCL bus bandwidth by up to 10% for operations using AllReduce, a distributed computing collective used to aggregate and synchronize data across every worker GPU. Oracle's benchmarking tests also recorded just over 14% for AllToAll operations, where every...
Keep reading with a free account
The rest of this article, and every signal for Oracle, is in your free account.
