Paving the Way Toward Energy-Aware and Automated Datacentre

Abstract

Energy efficiency and datacentre automation are critical targets of the research and deployment agenda of CINECA and its research partners in the Energy Efficient System Laboratory of the University of Bologna and the Integrated System Laboratory in ETH Zurich. In this manuscript, we present the primary outcomes of the research conducted in this domain and under the umbrella of several European, National and Private funding schemes. These outcomes consist of: (i) the ExaMon scalable, flexible, holistic monitoring framework, which is capable of ingesting 70GB/day of telemetry data of the entire CINECA datacentre and link this data with machine learning and artificial intelligence techniques and tools. (ii) The exploitation of ExaMon to evaluates the viability of machinelearning based job scheduling, power prediction and deep-learning based anomaly detection of compute nodes. (iii) The viability of scalable, out-of-band and high-frequency power monitoring in compute nodes, by leveraging low cost and open source embedded hardware and edge-computing, namely DiG. (iv) Finally, the viability of run time library to exploit communication regions in large-scale application to reduce the energy consumption without impairing the execution time, namely COUNTDOWN.

ANDREAS: Artificial intelligence traiNing scheDuler foR accElerAted resource clusterS

“…The profiling analysis described in this section has been performed to measure the power consumption of ARMIDA nodes and GPUs, as well as the expected execution times of the considered applications on all the configurations available in the reference hardware system. The profiling has been performed through ExaMon (Exascale Monitoring) a highly scalable framework for the performance and energy monitoring of HPC servers [2].…”

pAElla: Edge AI-Based Real-Time Malware Detection in Data Centers

“…The system is interfaced with the power sensor via a 12-bit 8-channels SAR ADC, and with existing in-band / out-of-band telemetries to collect hardware performance counters (e.g., Amester [39], IPMI [40], and RAPL [17]). Moreover, it includes (i) hardware support for the Precision Time Protocol (PTP), which allows sub-microsecond measurements synchronization [43], [44], (ii) two Programmable Real-Time Units (PRU0 and PRU1), that we exploit for real-time feature extraction on-board, and (iii) an ARM Cortex-A8 processor with NEON technology, useful for DSP processing and edge ML inference (e.g., by leveraging the ARM NN SDK [45], which enables efficient translation of existing NN frameworks -such as TensorFlow -to ARM Cortex-A CPUs); • a scalable distributed data analytics framework, namely ExaMon [15], [16], that we use to collect at a lower rate -from seconds to milliseconds -power and performance activity of the all cluster and thus to carry out cluster-level analytics. To send data from the distributed monitoring agents (i.e., daemons running on the BBBs) to the centralized monitoring unit, we adopted MQTT [46], which is a robust, lightweight and scalable publish-subscribe protocol, already used for large-scale systems both in industry and academia (e.g., Amazon, Facebook, [46], [47]).…”

“…For the performance metrics, we use a time granularity of 20 ms i.e., in line with SoA DCs / SCs monitoring tools [15]), dedicating 2 cores for ExaMon to prevent that the in-band monitoring can affect the measurements with noise. For this purpose, we exploit isolcpus, which is a kernel parameter that can be set from the boot loader configurations.…”

“…This is true also in the context of Supercomputers (SCs) and Data Centers (DCs), where a new generation of high-resolution monitoring systems is being deployed [10]- [12], pushing the research boundaries over new opportunities to support their automation, analytics, and control. In particular, DiG [10] is the first out-of-band monitoring system for DCs / SCs that allows real-time edge intelligence on high-resolution measurements, and is already installed in a SC in production (i.e., D.A.V.I.D.E., ranked 18 th most efficient SC in the world, based on Green500 November 2017 [13]- [15]). DiG is based on low-cost IoT embedded computers, which are essential to interface with analog power sensors to collect very fine-grain power measurements of the DC servers (i.e., up to 20 µs).…”

Monte Cimone: Paving the Road for the First Generation of RISC-V High-Performance Computers

“…When running an unoptimized Stream benchmark, Monte Cimone obtained just the 15.5% of the peak bandwidth, while Marconi100 and Armida obtained an efficiency of 48.2% and 63.21% respectively, pointing to significant margins for improvement in application and software stack tuning to the hardware target. • We extended the ExaMon monitoring framework [12] to monitor the Monte Cimone cluster. We characterised the power consumption of various applications executed on Monte Cimone.…”