Elements of Research Computing¶
DRAFT
The goal of this document is to introduce some general concepts, tools and best practices for research oriented computing. Advanced documents for specific topics can be arranged according to demand. Authors/speakers for specific topics are also welcome.
Preliminary remarks¶
The categorization "Research Computing" was chosen because, on one hand, that's the audience of this document, researchers that use computational tools; on the other hand, traditional terms like HPC, Grid, Cloud tend to separate from the onset the infrastructure that is going to be used to solve a specific problem, but in most cases one doesn't know what the best infrastructure(s) will be.
In any case, not all aspects of Research Computing will be covered - we will focus mainly on non-interactive "jobs" that have some sort of intensive requirement such as cpu, memory, network, storage, etc.
Some terminology¶
- High Performance
- perform a specific task in a short period of time (e.g. low latency)
- High Throughput
- perform many tasks in a fixed period of time (e.g. high bandwidth)
- Concurrent
- concurrency is a property of the algorithm (e.g. independence of tasks)
- Parallel
- concurrent parts of an algorithm can (or not) be run in parallel
- Distributed
- distributed generally means loosely parallel (e.g. asynchronous)
- Grid
- Grid usally means a collection of clusters with interoperability at scheduler level
- Cloud
- Cloud means a lot of different things (e.g. Infraestructure/Platform/Software as Services)