Design of an Energy Aware Petaflops Class High Performance Cluster Based on Power Architecture

Abstract

In this paper we present D.A.V.I.D.E. (Development for an Added Value Infrastructure Designed in Europe), an innovative and energy efficient High Performance Computing cluster designed by E4 Computer Engineering for PRACE (Partnership for Advanced Computing in Europe). D.A.V.I.D.E. is built using best-in-class components (IBM's POWER8-NVLink CPUs, NVIDIA TESLA P100 GPUs, Mellanox InfiniBand EDR 100 Gb/s networking) plus custom hardware and an innovative system middleware software. D.A.V.I.D. E. features (i) a dedicated power monitor interface, built around the BeagleBone Black Board that allows high frequency sampling directly from the power backplane and scalable integration with the internal node telemetry and system level power management software; (ii) a custom-built chassis, based on OpenRack form factor, and liquid cooling that allows the system to be used in modern, energy efficient, datacenter; (iii) software components designed for enabling fine grain power monitoring, power management (i.e. power capping and energy aware job scheduling) and application power profiling, based on dedicated machine learning components. Software APIs are offered to developers and users to tune the computing node performance and power consumption around on the application requirements.The first pilot system that we will deploy at the beginning of 2017, will demonstrate key HPC applications from different fields ported and optimized for this innovative platform.

Anomaly Detection Using Autoencoders in High Performance Computing Systems

“…Our target system is D.A.V.I.D.E. [ABBea17], an energy efficient supercomputer developed by E4 Computer Engineering [e4] and hosted by CINECA in Bologna, Italy. It is composed by 45 nodes with a total peak performance of 990 TFlops and an estimated power consumption of less than 2 kW per node.…”

Performance and Power Analysis of HPC Workloads on Heterogeneous Multi-Node Clusters

“…Furthermore, this latter limitation could be solved monitoring by also monitoring the wall socket power drain with supplementary external meters, or through motherboard embedded meters where available. Fortunately, hardware manufacturers and integrators are already moving along this direction [15], and also at the data center level; in the near future, a high amount of power related metrics should be made available to HPC users [44], also allowing for better support to the kind of application profiling we presented in this work.…”

“…Another promising approach towards energy-efficiency in HPC is the one of the EEHPCWG of N. Bates [13], trying to push an awareness action at the data center level. Part of this effort is the approach of optimizing the job scheduling using different power-aware policies presented by D. Tafani et al [14] or adding hardware/software extensions for improving energy awareness as presented by W. A. Ahmad et al [15]. Various attempts to take advantage of mobile technology for increasing energy efficiency of HPC systems have also been taken in the recent past.…”

Evaluation of NTP/PTP fine-grain synchronization performance in HPC clusters

“…In this paper we focus on the evaluation of NTP and PTP synchronization performance in a SoA HPC Cluster -i.e., D.A.V.I.D.E. [2] (18 th in Green500 November 2017) -and its out-of-band monitoring infrastructure, namely DiG and Examon [3,16]. A key concept in our study is that synchronization is performed within the LAN of the HPC infrastructure.…”