The automation of Network Services (NS) consisting of virtual functions connected through a multilayer packet-overoptical network requires predictable Quality of Service (QoS) performance, measured in terms of throughput and latency, to allow making proactive decisions. QoS is typically guaranteed by overprovisioning capacity dedicated to the NS, which increases costs for customers and network operators, especially when the traffic generated by the users and/or the virtual functions highly varies over the time. This paper presents the PILOT methodology for modeling the performance of connectivity services during commissioning testing in terms of throughput and latency. Benefits are double: first, an accurate perconnection model allows operators to better operate their networks and reduce the need for overprovisioning; and second, customers can tune their applications to the performance characteristics of the connectivity. PILOT runs in a sandbox domain and constructs a scenario where an efficient traffic flow simulation environment, based on the CURSA-SQ model, is used to generate large amounts of data for Machine Learning (ML) model training and validation. The simulation scenario is tuned using real measurements of the connection (including throughput and latency) obtained from a set of active probes in the operator network. PILOT has been experimentally validated on a distributed testbed connecting UPC and Telefónica premises.
Esta es la versión de autor del artículo publicado en: This is an author produced version of a paper published in: Multi-Gb/s links are becoming widespread, and network devices are under a continuous stress, so testing whether they guarantee the specified throughput or delay is a must. Software-based solutions, such as the packet-train traffic injection, were adequate for lower speeds, but they have turned inaccurate in the current scenario.Hardware-based solutions have proved to be very accurate, but usually at the expense of much higher development and acquisition costs. Fortunately, the new affordable FPGA SoC devices, as well as highlevel synthesis tools, can very efficiently reduce these costs. In this paper we show the advantages of hardware-based solutions in terms of accuracy, comparing the results obtained in an FPGA SoC development platform and in NetFPGA-10G to those of software. Results show that a hardware-based solution is significantly better, especially at 10 Gb/s. By leveraging high-level synthesis and open source platforms, prototypes were quickly developed. Noticeable advantages of our proposal are the high accuracy, the competitive cost with respect to the software counterpart, which runs in high-end off-the-shelf workstations, and the capability to easily evolve to upcoming 40 Gb/s and 100 Gb/s networks.
Customized compute acceleration in the datacenter is key to the wider roll-out of applications based on deep neural network (DNN) inference. In this article, we investigate how to maximize the performance and scalability of field-programmable gate array (FPGA)-based pipeline dataflow DNN inference accelerators (DFAs) automatically on computing infrastructures consisting of multi-die, network-connected FPGAs. We present Elastic-DF, a novel resource partitioning tool and associated FPGA runtime infrastructure that integrates with the DNN compiler FINN. Elastic-DF allocates FPGA resources to DNN layers and layers to individual FPGA dies to maximize the total performance of the multi-FPGA system. In the resulting Elastic-DF mapping, the accelerator may be instantiated multiple times, and each instance may be segmented across multiple FPGAs transparently, whereby the segments communicate peer-to-peer through 100 Gbps Ethernet FPGA infrastructure, without host involvement. When applied to ResNet-50, Elastic-DF provides a 44% latency decrease on Alveo U280. For MobileNetV1 on Alveo U200 and U280, Elastic-DF enables a 78% throughput increase, eliminating the performance difference between these cards and the larger Alveo U250. Elastic-DF also increases operating frequency in all our experiments, on average by over 20%. Elastic-DF therefore increases performance portability between different sizes of FPGA and increases the critical throughput per cost metric of datacenter inference.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.