Software Engineering Manager, Intel
Программно-определяемая инфраструктура: современные тренды в HPC
Modern high-performance computing can be divided into two big chunks: petaflops grade computing and petabytes scale analytics. Supercomputers are targeting to run programs in hours or days that would require decades or centuries on normal machine and designed for numerically-intensive applications. Petabytes scale internet datacenters, or so-called warehouse scale computers are supporting millions of customers, with mostly small transactions and large-scale analytics and designed for data collection, storage, and analysis. There is a desire to converge these two worlds into hyperscale superanalytical engine, to solve problems like large scale data driven modelling and simulation in real-time. Existing programming models are not well suitable for such future systems. One of the answers would be to have adaptable computing platform to ease development, deployment and run of the application. Modern approaches, tools and methods to build such Platform-as-a-Service will be discussed in the presentation.