Glossary¶
New to high-performance computing? This page explains the terms you'll meet most often across the manual. Terms are grouped by theme; acronyms also appear as hover tooltips throughout the rest of the site.
The basics¶
- CRCD Ecosystem
- The complement of CRCD infrastructure, including hardware, software, domain experts and services
- Resource Allocation
- An allotment of computing time and/or data storage quota
- HPC (High-Performance Computing)
- Using many connected computers together to run computational work that is too large or slow for a single desktop or laptop.
- Cluster
- A large collection of connected computers ("nodes") that work as one system. CRCD runs several clusters tuned for different workloads — see Hardware Profiles.
- Node
- A single computer within a cluster. Nodes come in two kinds below: login nodes and compute nodes.
- Login node
- The gateway machine you land on when you connect. Use it to edit files, manage data, and submit jobs — not to run heavy computation. See Login Nodes.
- Compute node
- A worker machine where your jobs actually run. You don't log into these directly; the scheduler assigns them to your job.
Hardware¶
- Core (CPU core)
- A single processing unit within a node. A job asking for "16 cores" reserves 16 of a node's cores to run in parallel.
- Thread
- A stream of execution within a program. Many programs run one thread per core; some run several.
- GPU (Graphics Processing Unit)
- A specialized processor that accelerates AI/ML and certain simulations. See the GPU cluster.
- RAM (memory)
- The fast working memory a job uses while running, requested separately from cores. Running out of memory is a common cause of failed jobs.
The CRCD clusters¶
- SMP cluster
- Symmetric Multi-Processing — for jobs that run on a single node and share one pool of memory. See SMP.
- MPI cluster
- For tightly coupled jobs that span multiple nodes using the Message Passing Interface. See MPI.
- HTC cluster
- High-Throughput Computing — for many independent single-node jobs, widely used for genomics and health-sciences work. See HTC.
- GPU cluster
- Nodes equipped with GPUs for accelerated computing. See GPU.
- TEACH cluster
- Resources set aside for courses and teaching. See TEACH.
- VIZ
- A graphical Linux desktop environment for visualization and GUI applications. See VIZ.
Jobs and scheduling¶
- Job
- A unit of work you submit to a cluster — a script plus the resources it needs (cores, memory, time).
- Batch job
- A job that runs unattended: you submit a script and the scheduler runs it when resources are free. Most work runs this way. See Slurm Batch Jobs.
- Interactive job
- A session on a compute node where you type commands live, useful for testing and exploration. See Interactive Jobs.
- Scheduler
- The software that decides which jobs run where and when, balancing everyone's requests fairly.
- Slurm
- The specific scheduler ("workload manager") CRCD uses. You interact with it
through commands like
sbatch,squeue, andscontrol. See the Slurm Overview. - Partition
- A named subset of a cluster's nodes that jobs are submitted to, often grouped by hardware type or purpose.
- Preemptible Partition
- A low-priority partition that runs on idle resources at no service-unit cost, but whose jobs can be preempted (interrupted) when higher-priority work needs the nodes. See Preemptible Partitions.
- Walltime
- The real-world clock time a job is allowed to run before the scheduler stops it. Request enough, but over-requesting can delay when your job starts.
- Allocation
- The share of computing resources granted to your research group, which your jobs draw from.
- Service Unit (SU)
- The accounting unit used to measure and charge cluster usage (roughly, one core running for one hour). See Service Units.
Software¶
- Module
- A pre-installed software package you load into your environment on demand
with
module load, rather than installing it yourself. See Discovering Software. - Lmod
- The tool that manages modules, letting you search (
module spider), load, and unload software versions cleanly. - Spack
- The build system CRCD uses to install and organize software on the current (RHEL 9) clusters, which is why modules are named the way they are.
- CRC wrappers
- CRCD's
crc-*helper commands (for examplecrc-interactive) that simplify common Slurm tasks. See CRC Wrappers.
Storage¶
- Home directory (
ihome) - Your personal space on the clusters (under
/ihome/), shared across all nodes and backed up. Good for code and small files, not large I/O. See File Systems. - Scratch space
- Fast temporary storage for the heavy read/write of a running job. Not backed up and periodically cleared, so move results out afterward. See Utilizing Scratch Space.
Getting connected¶
- PittNet
- the internal University network
Firewall :A network security device that monitors and filters incoming and outgoing network traffic based on an organization's previously established security policies
- VPN (Virtual Private Network)
- A secure network connection (Pitt's is GlobalProtect) required to reach the clusters from off campus. See Getting an Account.
- SSH (Secure Shell)
- The protocol for connecting to a login node from a terminal. See SSH using a terminal and Passwordless SSH.
- Open OnDemand (OOD)
- A web portal for using the clusters through your browser — files, notebooks, and interactive apps with nothing to install. See Open OnDemand.