RoCE

Definition

RoCE (RDMA over Converged Ethernet) enables Remote Direct Memory Access over standard Ethernet networks. RoCE v1 operates at the link layer (similar to InfiniBand), while RoCE v2 runs over UDP/IP, enabling routing across subnets. RoCE provides a lower-cost alternative to InfiniBand for RDMA-capable storage (NVMe-oF) and HPC interconnects, though it requires lossless Ethernet configuration (PFC/ECN) for reliable operation.

Written by
Daniel Kovacs

Updated Feb 11, 2026

Ansible

An open-source IT automation tool for configuration management and orchestration, widely used for HPC cluster and data center provisioning.

BeeGFS

A high-performance parallel file system with distributed metadata, simple administration, and built-in high availability for HPC and AI workloads.

Burst Buffer

A fast intermediate storage layer using NVMe SSDs that absorbs I/O bursts between compute nodes and the parallel file system.

Ceph

An open-source distributed storage platform providing unified object, block, and file storage with automatic data distribution and self-healing capabilities.

CXL

Compute Express Link â€” an open interconnect standard enabling high-speed communication and memory sharing between CPUs, accelerators, and memory devices.

Data Lake

A centralized repository storing raw data in its native format at scale, supporting diverse analytics workloads from batch to real-time processing.