Glossary

API (Application Programming Interface)

A set of functions you use to control a specific service.

Batch or Batch processing

Non interactive program, submitted to the scheduler, equivalent to a job.

Cluster

Collective organization of multiple and cohesively managed nodes (workstations or servers) combined to perform computing tasks that can be processed in parallel.

Embarrassingly parallel

This is the perfect parallel application, there is no effort to separate the workload into a number of parallel and independent tasks. In general, the speed up is linear.

Head node or master node

This is the machine on which the user logs in practice when he needs to connect to the cluster, this machine is used for submitting jobs and handling the management tasks.

HPC platform

The HPC platform contains a head node machine.

• It contains nodes from 10 to 100k.

• A node can contains multiple sockets, i.e processors from 1 to 8.

• A socket can contain multiple cores (1, 4, 8,...)

Job

A job is a request addressed to the scheduler for running task(s) using specified resource(s), it is defined in a job script containing all the information needed for its execution.

Job ID

The job scheduler assigns to each job a unique ID. This ID is used to check the status of the job or to manage it (cancel, change of associated priority, …).

Master/Worker

This is a well-known parallel model where a master manages and distributes tasks to slaves. Simcenter Amesim batch and design exploration studies are master/worker applications.

Moore's law

Today, this is the only way to increase the power of computation as processors reach their performance limits, CPU clock speed has not progressed for years. For this reason, performance improvement comes exclusively from additional resources and parallel computing.

Node

Machine (server or workstation) used to process a job, allocated to the job.

Run

A run is a unitary Simcenter Amesim model simulation: it can be a single run or one unitary run of an Simcenter Amesim study such as a batch run or one simulation step of a gradient computation.

Note

A Simcenter Amesim run itself cannot be parallelized; only several independent runs can be.

Scalability

Ability to improve the performance with additional resources.

Scheduler or Job scheduler

Application running on the HPC platform for managing the workload of jobs to ensure their successful completion under shared and global priority rules.

Scheduling or Scheduling strategy

Various schemes are used to manage the job queue, considering computer resources, job priority, user profile, ...

Speed up / linear speed up

Speedup is usually discussed as a function of the number of processors, to evaluate the quantity of time saved thanks to additional resources. Linear speedup means the time gain is proportional to the amount of resources added.

Study

A "study" denotes any kind of Simcenter Amesim simulation among single runs, batch runs or design exploration studies (Monte Carlo analysis, Design of Experiments or Optimization).