|
Background
Workflow
Cluster Operations
edit SideBar
|
General Terms
- client
- the user's (your) terminal accessing cluster resources
- CPU
- central processing unit, performs calculations by moving data in and out of memory sequentially (slower)
- GPU
- graphical processing unit, performs calculations by moving data in and out of memory in chunks (faster)
- parallel processing/multithreading
- splitting a task into independent smaller tasks to be distributed across multiple CPUs/GPUs
- cluster
- a network of computers with distinct purposes (login, storage, and compute nodes - see below)
- core
- 1 CPU processor
- node
- 1 computer on the cluster, a collection of CPUs and GPUs (n78, n90, etc)
- head node
- the computer used for fast/infrequent requests like logins, file manipulation, and running the scheduler (cottontail2)
- queue
- a collection of similar nodes (mwgpu, exx96, etc)
- scheduler
- slurm, a program that reads header information (#SBATCH lines) from job scripts and dispatches the calculations to compute nodes
List of Useful Nodes
- cottontail2
- head node
- n33-n37
- compute nodes on the mwgpu queue
- n79-n90
- compute nodes on the exx96 queue
|
