In this deliverable, we explore this question by studying accelerator deployment models. Under accelerator, we understand for example application-specific GPUs or specially programmed FPGAs. A deployment specifies types, amount, and connectivity of accelerators in a datacenter. With these definitions in mind, we created a theoretical model of the datacenter, its components, expected workloads, and finally, it is possible deployments.
We have developed VineSim, a software simulator of a datacenter, based on the aforementioned theoretical modeling. VineSim takes as inputs a workload and a deployment description and outputs performance metrics of interest, such as job latency and resource utilization. In VineSim, one can configure several parameters, including how tasks are allocated to nodes, and estimations of how fast they execute on different accelerators. VineSim can be used to explore how different deployments respond to different kinds of workloads, thus allowing one to determine how to best compose a datacenter based on particular workload, performance, or budgeting requirements.
This document presents the executive summary of the deliverable on Programming Language and Runtime System: Early Prototype, which aims at describing the core functionality of the VINEYARD programming model and runtime system for accelerated data centres. We describe our approach to creating an abstract representation of accelerated kernels, such that application programmers can use these kernels without needing to worry about accelerator-specific calling conventions, or about the specific versions available in the VINEYARD accelerator library.
The second key contribution of this document is the description of our approach to virtualizing accelerators. We assume that accelerators are assigned to jobs only when they are really needed, and not at job allocation time. This raises issues that need to be addressed in the virtualization layer and also in the application’s runtime. We describe these issues and our approach to solving.
The VINEYARD projects aims to achieve easy-to-use and transparent acceleration of data analytics. One of the components in the VINEYARD is the programming model and runtime system support, which is developed in Work Package 4. This document elaborates the requirements for the VINEYARD programming model and runtime system.
This document presents part of the deliverable on system architecture which aims at describing the overall architecture of VINEYARD and specifically the hardware and the software components that are developed in VINEYARD. VINEYARD’s goal is to both develop energy efficient hardware-accelerated servers and to develop the required framework for the seamless utilization of these servers in the programming frameworks that are widely used by the applications developers.
To this end, this document describes the VINEYARD platform and the VINEYARD framework. The VINEYARD platform consists of the hardware devices that are used and developed during the project while the VINEYARD framework consists of all the software, middleware, APIs, libraries and GUIs that are developed for the efficient integration of the hardware platforms.
This document presents the Executive Summary of the deliverable "Workload & Traffic Pattern Characterization". In the deliverable, it is discussed the characteristics of applications that we expect will benefit from the use of accelerators.
Power Point presentation of VINEYARD project