D5.1: Accelerator Deployment Models

Eleni Kanellou, Nikolaos Chrysos, Angelos Bilas, Christoforos Kachris
2017 Zenodo  
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
more » ... r, 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.
doi:10.5281/zenodo.898170 fatcat:l6gfsswmbbg7niupwnix7tfmjm