An HPC Glossary

If you don't find our definitions satisfying, or can't find what you want below, then try wikipedia or Foldoc 

  • batch system - A system that takes a set of commands (a job) and executes it without further intervention from the user. Normally a job would require input from one or more files and send output back to other files.  There is usually a queuing system that allocates jobs to the available resources via a scheduler.
  • directory -  equivalent of  a Windows folder
  • distributed-memory - memory which is not local to the node and can only be accessed by sending a message to another node. Distributed-memory is slower than shared-memory and often hardeer to program but ultimately it scales too much larger problems.
  • flag - An argument to a command that modifies how it works.
  • home directory - the directory you see when you first login.
  • HPC - originally High Performance Computing, but High Productivity Computing describes what we aiming for much better. The aim is to vastly improve on what you can achieve with your desktop.
  • MPI - Stands for Message-Passing Interface - it supports a parallel programming method designed for distributed-memory, but can be used very well on shared-memory stems also.
  • node - a single computer in a cluster. This may have multiple processor-cores which share it's own local memory.
  • shared-memory - Memory which is not local to the processor but can be accessed directly and fairly rapidly without the need for messages to be sent to the other processor. The memory on an individual node is usually shared-memory.
  • shell - A software layer between the user and the operating sytem that interprets the commands a user types.