Glossary

This glossary defines key terms used throughout the Omnia documentation. Terms are listed alphabetically. Where applicable, entries link to the documentation page that provides a full explanation.

Apptainer

Formerly known as Singularity. A container runtime designed for HPC environments that allows users to run containers without root privileges. Supported on Slurm compute nodes in Omnia 2.1+.

BMC

Baseboard Management Controller. A dedicated microcontroller embedded in server motherboards that provides out-of-band management capabilities (power control, hardware monitoring, remote console) independent of the host operating system. On Dell PowerEdge servers, the BMC is implemented as iDRAC.

BSS

Boot Script Service. A component of OpenCHAMI that dynamically generates per-node boot scripts based on the node's hardware profile and assigned role. BSS provides boot configuration during network provisioning.

BuildStreaM

Omnia's GitLab CI-based automation pipeline for catalog-driven deployment. Administrators define a deployment catalog (YAML manifest), and BuildStreaM generates a CI/CD pipeline that executes the required Ansible playbooks in order. See Components.

Calico

A CNI (Container Network Interface) plugin for Kubernetes that provides pod-to-pod networking and network policy enforcement. Omnia deploys Calico as the default CNI in the Kubernetes cluster.

cloud-init

A cloud instance initialization system that configures systems during first boot. Omnia uses cloud-init to automate node configuration and customization during the provisioning process.

Composable Roles

Omnia's system for assigning server functions via a declarative mapping file. A single server can hold multiple roles (e.g., Slurm control node + login node), decoupling physical hardware from logical cluster functions.

Containerd

An industry-standard container runtime with an emphasis on simplicity, robustness, and portability. Omnia uses Containerd as the container runtime for Kubernetes workloads.

CoreDHCP

A DHCP server implementation that provides network boot and configuration services. Omnia uses CoreDHCP for PXE boot provisioning and multi-subnet DHCP configuration for rack-based deployments.

CoreDNS

A DNS server that is flexible and extensible, serving as the foundation for service discovery in Kubernetes. Omnia uses CoreDNS for dynamic DNS resolution and hostname management via coresmd.

CSI drivers

Container Storage Interface drivers that enable Kubernetes to orchestrate storage from arbitrary storage systems. Omnia deploys CSI drivers for PowerScale and other storage backends.

CUDA

NVIDIA's parallel computing platform and programming model. Omnia automates CUDA toolkit installation for GPU-enabled nodes to support HPC and AI workloads.

DCGM

Data Center GPU Manager. NVIDIA's suite of tools for monitoring and managing GPU data center environments. Omnia deploys DCGM for GPU telemetry collection and monitoring.

DOCA-OFED

NVIDIA's data center acceleration on Arm and NVIDIA OpenFabrics Enterprise Distribution. Omnia automatically installs DOCA-OFED drivers for Mellanox InfiniBand adapters to enable high-performance networking.

ETCD

A distributed, reliable key-value store for the most critical data of a distributed system. Omnia uses ETCD as the Kubernetes cluster state store, with support for local disk deployment for high availability.

Functional Groups

Named role definitions in Omnia's Composable Roles system. Each functional group (e.g., slurm_node, service_kube_control_plane) determines which software and configuration is applied to a server.

iDRAC

Integrated Dell Remote Access Controller. Dell's implementation of the BMC, providing Redfish API access, remote console, virtual media, firmware management, and hardware telemetry for Dell PowerEdge servers. See Telemetry Architecture.

InfiniBand

A high-performance, low-latency networking technology designed for HPC clusters. Omnia supports InfiniBand networking with automatic DOCA-OFED driver installation for Mellanox adapters.

iSCSI

Internet Small Computer System Interface. A storage networking protocol for linking data storage facilities. Omnia uses iSCSI for PowerVault storage integration to provide persistent storage for critical cluster components.

Kafka

Apache Kafka. A distributed event-streaming platform used as the central message broker in Omnia's telemetry pipeline. All metrics from iDRAC and LDMS flow through Kafka before reaching VictoriaMetrics. See Telemetry Architecture.

LDAP

Lightweight Directory Access Protocol. A protocol for accessing and maintaining distributed directory information services. Omnia integrates LDAP for centralized authentication and user management.

LDMS

Lightweight Distributed Metric Service. A high-performance, low-overhead metric collection framework developed by Sandia National Laboratories for HPC environments. Omnia deploys LDMS agents on compute nodes to collect in-band OS and application metrics. See Telemetry Architecture.

MetalLB

A bare-metal load balancer for Kubernetes. MetalLB assigns external IP addresses to Kubernetes LoadBalancer services in environments without a cloud provider. Omnia deploys MetalLB automatically in the Kubernetes cluster.

OIM

Omnia Infrastructure Manager. The dedicated management node that runs all Omnia control-plane services, including the omnia_core Ansible container, OpenCHAMI, Pulp, and telemetry collectors. The OIM is the single point from which the entire cluster is provisioned and managed. See Architecture.

OpenCHAMI

Composable Hierarchical Automated Management Infrastructure. The provisioning engine at the core of Omnia, providing API-driven node discovery, hardware inventory, and bare-metal lifecycle management. Includes SMD and BSS as sub-services. See Components.

OpenManage Enterprise (OME)

Dell's infrastructure management solution that provides unified management for Dell PowerEdge servers. Omnia integrates with OME for automated BMC discovery and inventory management.

OpenTelemetry

An observability framework for cloud-native software. Omnia uses OpenTelemetry Collector for PowerScale telemetry collection and data transformation.

Podman

A daemonless, rootless container engine for running OCI containers on Linux. Omnia uses Podman on the OIM to run all management services (omnia_core, OpenCHAMI, Pulp, OpenLDAP) as isolated containers without requiring Docker. See Architecture.

PowerScale

Dell's enterprise-scale network-attached storage (NAS) platform. Omnia integrates with PowerScale for storage provisioning, telemetry collection, and CSI driver support.

PowerVault

Dell's direct-attached storage (DAS) and network-attached storage (NAS) solutions. Omnia integrates with PowerVault for persistent storage using iSCSI block storage with multipath support for Slurm controller components.

PXE

Preboot Execution Environment. An industry-standard protocol that allows servers to boot an operating system image over the network rather than from local disk. Omnia uses PXE for initial node discovery and OS provisioning.

Pulp

An open-source repository management platform. Omnia deploys Pulp as a Podman container on the OIM to mirror RPM repositories, container images, and Python packages locally. Essential for air-gapped deployments. See Components.

ROCm

Radeon Open Compute. AMD's open-source software platform for GPU-accelerated computing. Omnia supports ROCm installation on nodes with AMD Instinct GPUs for AI/ML and HPC workloads.

SMD

State Manager Daemon. The inventory and state-tracking service within OpenCHAMI. SMD maintains a real-time record of every node's hardware configuration, power state, and component hierarchy.

SmartFabric Manager (SFM)

Dell's network management solution for SONiC that streamlines network management with automation, analytics, and scalability. Omnia supports SFM telemetry collection for fabric monitoring and performance analysis.

Slurm

Simple Linux Utility for Resource Management. An open-source, highly scalable job scheduler and workload manager widely used in HPC clusters. Omnia deploys and configures Slurm for batch job scheduling on compute nodes. See Architecture.

TFTP

Trivial File Transfer Protocol. A simple file transfer protocol used during PXE boot to deliver the initial bootloader binary to bare-metal nodes.

Vector

A high-performance, vendor-neutral observability data pipeline. Omnia uses Vector for collecting, transforming, and routing telemetry data from LDMS and OME sources to VictoriaMetrics and VictoriaLogs.

VictoriaLogs

A high-performance log database built for log storage and analysis. Omnia uses VictoriaLogs for centralized log collection and storage, working alongside VictoriaMetrics for complete observability.

VictoriaMetrics

A high-performance time-series database that stores all telemetry data in Omnia's monitoring pipeline. Supports PromQL and MetricsQL for querying, and achieves high compression ratios for efficient long-term metric retention. See Telemetry Architecture.

vlagent

VictoriaLogs agent that collects logs from various sources and pushes them to VictoriaLogs. Omnia deploys vlagent for log collection from cluster components.

vminsert

VictoriaMetrics component responsible for ingesting time-series data into the VictoriaMetrics database. Omnia uses vminsert as the data ingestion endpoint for metrics storage.

vmselect

VictoriaMetrics query component that executes queries against stored metrics data. Omnia uses vmselect for querying and retrieving metrics from the VictoriaMetrics database.

vmstorage

VictoriaMetrics storage backend component that persists time-series data. Omnia uses vmstorage for durable storage of metrics data in cluster mode deployments.

vmagent

VictoriaMetrics agent that collects metrics from various sources and pushes them to VictoriaMetrics. Omnia deploys vmagent for metrics collection from cluster components.