High-performance computers with high throughput, high utilization rate and low latency
High-throughput computers are a new type of high-performance computer that is suitable for emerging application loads such as those of the Internet big data. They can globally control the processing of high-concurrency requests under strong time constraints. Their core characteristics are the guarantee of concurrency, real-time, and determinism.
The characteristics of high-performance computing in traditional scientific and engineering computing applications include: single tasks, infrequent changes in load, large single-task computing volume, and good locality of computation. While high-throughput computing in data centers is mainly applied in emerging scenarios such as the Internet and Internet of Things, its characteristics are: diverse tasks, single tasks often having streaming computing features; relatively small computing volume, but a huge number of concurrent tasks and data scale; and processing requirements having real-time nature.
The development goal of traditional high-performance computers is to increase speed, that is, to shorten the running time of a single parallel computing task; while the goal of data center application systems is high throughput, that is, to increase the throughput of tasks or data processing per unit time. High-performance computers with this performance metric of “doing more computing” are called high-throughput computers.