Glossary

Peak achievable data transfer rate across a network link/path, measured in bytes per second

A set of possibly many computers (or compute nodes) interconnected via a high-speed network used as a high-capacity compute platform

A computer in a compute platform such as a cluster

A software service that manages compute resources and handles requests for executing computational tasks

A processing unit, or processor, in a multi-core microprocessor

The fraction of time that a cpu (core) spends computing during a program execution

The longest path, in seconds, from an entry task to an exit task in a directed acyclic graph (dag) of tasks.

A set of hardware resources (hardware infrastructure) managed by software services (software infrastructures) that make it possible to execute parallel programs

The concept of having date "close" to where the computation takes place. this could be having data in a cache close to the core withing a computer, or data in storage close to a computer (in terms of network proximity).

A directed graph containing no cycles, often used to represent a workflow

The time it takes for a task/program to execute from start to finish.

A standard unit of "amount of work to do" used for scientific applications (as many of such applications perform mostly floating point arithmetic); computational hardware resources then have speed measured in flop/sec, i.e., how much "work" they can perform per time unit

A pattern in a directed acyclic graph (workflow) in which one vertex (source task) is connected to many other vertices (children tasks), and these vertices are then all connected to a single vertex (sink task).

The entry-point computer to a cluster, which typically is connected to the internet and runs software to handle requests for the cluster's compute nodes

The amount of time that a core/computer does not actually compute during a program execution (typically because of i/o operations taking place)

The time it takes for a single packet to travel from one end of a network link/path to the other end

The overall execution time of a program execution (i.e., the wall-clock time between the beginning of the program's first task and the competion of the program's last task).

A graph in which edges are network links and vertices are either end-points (i.e., computers) or routers

The parallel speedup divided by the number of processors, as a percentage

The ability of a parallel program to maintain high parallel efficiency as the number of compute nodes / cores increases

The acceleration factor (when compared to a sequential execution) when running an application using multiple processors

The ability of a program to execute multiple streams of instructions simultaneously on multiple processors

A storage device, such as a hard drive, that retains data even after its power source has been shut off

Memory provided by a computer so that running programs can store all the code/data needed for their execution on that computer

A software service that manages persistent storage and handles requests for creating, deleting, reading, and writing files

The technique by which an operating system runs multiple programs on the same core, rapidly cycling over them throughout time.

The time elapsed between a task is submitted for execution and its completion

The number of operations that a particular task needs to execute when executed on a core that performs some number of such operations per time unit. for instance, a task with 100 gflop work would run in .5 seconds on a core with 200 gflop/sec speed.

A program comprised of individual computational tasks, often stand-alone executables, that must execute in some particular sequence to produce a final desired output

A software tool that automatically orchestrates the execution of a workflow on a cyber infrastructure