Home » HPC glossary
HPC glossary
A
Amdahl's law
It predicts the improvement of theoretical performance by changing the processors number. Wikipedia
B
Bandwidth
It indicates the performance of HPC systems components as memory, I/O and interconnection with latency periods. Wikipedia
Batch
Program performance characterised by the absence of the user interaction. Wikipedia
Beowulf
Linux based PC cluster realized by commodity hardware and open-source software use. Wikipedia
C
Cloud computing
Distributed computing paradigm where required computer ICT Resources are supplied as a "service" by the network. Cloud computing provides on demand a scalable computing power to users. Wikipedia
Cluster / Cluster computing
A cluster of computers is a supercomputer realized by connecting a group of PCs, servers and workstations in a unique network, which collaborate through a software to share the workload. The hardware of these systems can be made as simple computer network (they generally are compact servers located in a rack) and the "fusion" takes place only at the software level allowing the computers to cooperate as a unique computer multiprocessor node. Wikipedia
Clusters - High Performance Computing Clusters
Cluster where computing power of all available processors is used to reduce (or arrange) computing time of complex processing thanks to the involvment of parallel computing techniques.
Clusters - Load Balancing Clusters
They are cluster where the requests sent to a server (eg. web-server, database-server...) are distributed by load balancing algorithms to obtain high levels of performance, which is almost equivalent to the sum of the computing capacity of each computer. Wikipedia
Computing node
It is a generic node used to accomplish computing workload of a cluster.
Core
Modern CPU hosts multiple cores in their inner part and this is considered as an effective independent processors by the operative system and the application software. Wikipedia
D
Distributed filesystem
In a network file system, the files are memorized on networking devices (servers, PCs, storages...), and they are managed in client/server mode. More common filesystems are NFS and CIFS. Wikipedia
Distributed memory
Distributed memory refers to a multiple-processor computer system in which each processor has its own private memory. Wikipedia
E
Execution queues
An administrator of execution queues is a software that manages the workload of a HPC system. Users submit programs in batch queue, so they can be executed as soon as resources become available.
F
Flops
Flops is an acronym meaning FLoating point Operations Per Second and it is a measure of a computer performance. It expresses the number of mathematical operations between numbers in floating point per second. Wikipedia
Front-end
It is a cluster node used to interface the external world, it hosts software and network services required by the cluster, accounts and users data, software applications; it is excluded from computing workload of the cluster which is shared out between the nodes.
G
Gigabit Ethernet
It is an Ethernet standard version defined in 1998, relative to network connection increasing the speed to 1000 Mbit/s. It is the basic interconnecting system of modern clusters. Wikipedia
Grid computing
Distributed computing paradigm with interconnection and sharing resources (computers, workstations, clusters...) decentralised into unique scalable infrastructure that offers high computing power and storage. Wikipedia
H
High Availability (HA) clusters
This is a Cluster where the availability of independent computers is used for redundant service networks of "critical mission" kind, in order to reduce the "single point of failure" and maximize availability. Wikipedia
I
Interconnection
Since the required interconnection speed can limit complex performances for certain applications or if a node quantity increasing is noticed, it is possible to employ alternative specific technologies, mostly based on optical fibres, able to increase bandwidth and reduce latency periods (i.e. Infiniband).
L
Latency
Latency is the time between the beginning and the end of an operation. In data transmission context it is the time required by a data to arrive at its destination. It expresses, together with the bandwidth, the HPC system components performances, such as memory, I/O and interconnection. Wikipedia
Load balancing
Computing workload is evenly distributed into a set of parallel processors. Parallel programs reach maximum efficiency when the workload is perfectly balanced, i.e. when each processor has the same work to perform.
M
Master node
Term employed to indicate a node that simultaneously performs the role of computing node and front-end node in a cluster.
MPI (Message Passing Interface)
It is a standard software that describes a communication protocol for parallel applications based on the messages exchanging. It also defines a library of functions through which parallel processors communicate each other. Wikipedia
Multiprocessor
It is a computer with two or more processors operating in parallel with a unique shared memory. Wikipedia
Multitasking
It is a method by which multiple processes are executed by a single processor; it is generally obtained by scheduling the time of the processor, under the control of the operating system. Wikipedia
N
Node
It is the common word used to indicate a generic computer belonging to a cluster. Generally a node can host several processors and/or cores.
O
OpenMP
It is a standard software that describes a parallel programming system for shared memory computers. Unlike MPI, OpenMP operates on high level and simplify parallel programming, even if it offers a lower control. Wikipedia
P
Parallel computer
This is a computer supplied by many processors able to accomplish at different instructions simultaneously. Through parallel computing it is possible to exploit computing capacity of this computers. Wikipedia
Parallel computing
Parallel computing allows to split a program into different tasks each one working in cooperation on different processors to reduce overall computing time. Wikipedia
Parallel filesystem
It is a filesystem with the support for parallel applications, in which all nodes can simultaneously access to the same file, for reading and writing. Data are registered on each disk unit of the different nodes (striping), in order to obtain I/O high performance and scalability. The most common filesystems are Lustre, PVFS, GPFS, Panasas ActiveScale. Wikipedia
S
Scheduler
A scheduler distributes HPC system workload in a balanced and optimize way, depending on the resources required and available. Wikipedia
Shared memory
In parallel computers a shared memory offers a single memory space available to all the processors. Wikipedia
SMP
Acronym for Symmetric MultiProcessor, it indicates a computer with two or more processors and a single shared memory. Wikipedia
Speedup
It refers to the performance improvement in program execution using multiple processors in parallel computing compared to single processor use. Wikipedia
Supercomputer
The term supercomputer generally indicates a computer designed to obtain high performances. Historically it represents a kind of systems for High Performance Computing characterized by a dedicated and expensive hardware architecture, it was used before cluster arrival. Wikipedia
T
Top500.org
This is the ranking indicating the 500 most powerful known computer systems in the world. Wikipedia
W
Workstation
It is a high performance computer used in professional environment for activities which require high levels of performance and reliability. Wikipedia
Copyright © 2012 Aethia Srl - VAT N. 07979630014