CIQ Glossary: Clear Definitions for High Performance Computing, Linux, and AI

An accelerator card is specialized hardware that offloads and speeds up specific workloads, such as GPUs, FPGAs, and ML accelerators.

AI training is the compute-intensive process of teaching a model from data; inference is running the trained model to make predictions.

Batch processing runs large volumes of work automatically as scheduled batches, with no user interaction once a job is submitted.

A Beowulf cluster is a high-performance computing cluster built from commodity computers and open source software networked together to run work in parallel.

CentOS is a family of RHEL-derived Linux distributions; CentOS Linux was a free downstream RHEL clone, now succeeded by the upstream CentOS Stream.

Cloud bursting is a hybrid-cloud technique that automatically shifts overflow workload to public cloud resources when local capacity runs short.

A cluster is a set of computers linked so their resources can be pooled to work on computational jobs as a single system.

Computational fluid dynamics (CFD) is the numerical simulation of how liquids and gases flow, using computers to solve fluid-motion equations across a modeled space.

A computational task is a single unit of work in a computing process, often one step of a larger job that a system runs and schedules.

A compute node is a single server or cloud instance in a cluster that provides processing resources for computational jobs to run on.

A container image is a standalone, executable package containing an application's code, runtime, libraries, and settings, from which containers are created.

Containerization packages software with all its dependencies into an isolated, portable unit that runs consistently on any compatible host operating system.

Data science is the practice of analyzing large datasets computationally to extract insight, often using HPC for heavy analytics.

DNS (Domain Name System) is the internet's directory service that translates human-readable domain names into the IP addresses computers use to connect.

Electronic design automation (EDA) is the software used to design, simulate, and verify integrated circuits and electronic systems before they are manufactured.

An embarrassingly parallel workload splits into independent tasks that need little or no communication, so it scales across many cores easily.

Encryption key management is the practice of securely handling cryptographic keys across their full lifecycle, from generation to deletion.

EPEL is a community-maintained repository of extra open-source packages for RHEL and compatible distributions, beyond what the base OS provides.

Finite element analysis (FEA/FEM) is a numerical method that breaks a complex structure into small elements to simulate stress, heat, and other physical forces.

An FPGA is a reprogrammable chip of configurable logic gates that can be tailored in hardware to accelerate a specific workload.

Genomic sequencing determines the order of DNA base pairs in a biological sample, then computationally aligns those reads against a reference genome.

A GPU is a processor with thousands of small cores built for parallel computation, used for graphics, HPC, and AI training and inference.

High-throughput computing runs huge numbers of independent jobs across distributed resources, maximizing work completed over days and weeks.

High performance computing (HPC) uses many processors working in parallel to run calculations and analyze data far faster than a single computer.

HPC architecture is the design of computing systems that combine many processors, memory, storage, and interconnects to run large workloads in parallel.

An HPC data center is a facility purpose-built to power, cool, network, and house high performance computing clusters and their storage.

An HPC interconnect is the very high-speed network that moves data between compute nodes fast enough for tightly coupled parallel workloads to run efficiently.

Infrastructure as code (IaC) manages and provisions IT infrastructure through machine-readable definition files instead of manual configuration.

Interactive jobs run on cluster resources while a user works with them live through an interface, unlike batch jobs submitted to run unattended.

IT infrastructure is the hardware, software, networks, and services that together support the delivery, storage, and processing of information technology.

A Jupyter Notebook is a browser-based document that combines live code, its output, and explanatory text in one shareable file.

MPI is a standardized library for passing data between processes on different nodes, letting one program run across thousands of cores.

Parallel computer architecture is the hardware design in which many processors work together concurrently to solve a single problem faster.

Parallel computing splits a problem into smaller parts that run at the same time on many processors or nodes, cutting time to solution.

A parallel file system stores data across many storage servers so thousands of compute nodes can read and write simultaneously at high bandwidth.

A pipeline is a sequence of processing stages where the output of each stage feeds the next, used to process data or execute instructions.

Provisioning is the process of setting up and configuring IT resources, servers, networks, storage, and user access, so they are ready to use.

RDMA lets one computer read or write another's memory directly over the network, bypassing both operating systems for very low latency.

Research computing is the people and services, often a university department, that run the HPC clusters and tools researchers rely on.

RHEL forks are Linux distributions rebuilt from Red Hat Enterprise Linux source code to provide RHEL-compatible operating systems, often free of charge.

Rocky Linux is a free, community-driven Enterprise Linux distribution built to be bug-for-bug compatible with Red Hat Enterprise Linux (RHEL).

Role-based access control (RBAC) grants system permissions through roles tied to job functions rather than to individual users.

Samba is open-source software implementing the SMB protocol, letting Linux and Unix systems share files and printers with Windows clients.

Scaling out adds more nodes to a system to grow capacity, rather than making existing machines individually more powerful.

Scaling up, or vertical scaling, adds capacity by making existing machines more powerful rather than by adding more machines.

The HPC triangle is the balance of fast compute, fast storage, and fast networking that together determine a cluster's real-world performance.

The Linux kernel is the core of the Linux operating system, managing hardware, memory, processes, and the interface between userspace programs and devices.

Two-factor authentication (2FA) verifies identity with two independent factors, so a stolen password alone cannot grant access.

Userspace is the unprivileged part of a Linux system where user applications run, separated from the kernel and requesting services through system calls.

Video transcoding converts video from one format, codec, or resolution to another so it can play across different devices and networks.

Weather and climate modeling simulates the atmosphere on a 3D grid to forecast conditions and study long-term climate, one of the largest HPC workloads.

An HPC workload is a catch-all term for the kind of work run on a cluster, such as genomics, weather modeling, or AI training.

YUM (Yellowdog Updater, Modified) is a package manager for RPM-based Linux distributions that automates installing, updating, and resolving dependencies.