Elastic-DF: Scaling Performance of DNN Inference in FPGA Clouds through Automatic Partitioning

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.

ESM-Tools Version 4.0: A modular infrastructure for stand-alone and coupled Earth System Modelling (ESM)

Earth system and climate modelling involves the simulation of processes on a wide range of scales and within and across various components of the Earth system. In practice, component models are often developed independently by different research groups and then combined using a dedicated coupling software. This procedure not only leads to a strongly growing number of available versions of model components and coupled setups but also to model- and system-dependent ways of obtaining and operating them. Therefore, implementing these Earth System Models (ESMs) can be challenging and extremely time-consuming, especially for less experienced modellers, or scientists aiming to use different ESMs as in the case of inter-comparison projects. To assist researchers and modellers by reducing avoidable complexity, we developed the ESM-Tools software, which provides a standard way for downloading, configuring, compiling, running and monitoring different models - coupled ESMs and stand-alone models alike - on a variety of High-Performance Computing (HPC) systems. (The ESM-Tools are equally applicable and helpful for stand-alone as for coupled models. In fact, the ESM-Tools are used as standard compile and runtime infrastructure for FESOM2, and currently also applied for ECHAM and ICON standalone simulations. As coupled ESMs are technically the more challenging tasks, we will focus on coupled setups, always implying that stand-alone models can benefit in the same way.) With the ESM-Tools, the user is only required to provide a short script consisting of only the experiment specific definitions, while the software executes all the phases of a simulation in the correct order. The software, which is well documented and easy to install and use, currently supports four ocean models, three atmosphere models, two biogeochemistry models, an ice sheet model, an isostatic adjustment model, a hydrology model and a land-surface model. ESM-Tools has been entirely re-coded in a high-level programming language (Python) and provides researchers with an even more user-friendly interface for Earth system modelling lately. The ESM-Tools were developed within the framework of the project Advanced Earth System Model Capacity, supported by the Helmholtz Association.

Craig Interpolation for the Integers: Results, Implementation, and Experiences

Craig interpolation is a versatile tool in formal verification, in particular for generating intermediate assertions in safety analysis and model checking. Over the last years, a variety of interpolation procedures for linear integer arithmetic (and extensions) have been developed. I will give an overview of the existing algorithms and design choices, and then discuss implementations of such procedures within theorem provers and SMT solvers. In particular, I will describe an implementation done using the multi-paradigm language Scala, which is built on top of the Java runtime infrastructure, and evaluate performance and engineering aspects.

Technical Note on Secure Service Access in VANET using PBKDF Algorithm

Vehicular networks and their applications have gained great attention to the research community and vehicle industry in past few years. The applications in Vehicular Ad-hoc Networks (VANETs) are growing rapidly. The two main application classes have lately gained popularity such as secured and non secured applications. In this paper, we focus on security in VANETs using Packet Bit Key Data Function (PBKDF) Algorithm. The performance of proposed algorithm is assessed using Simulation of Urban Mobility (SUMO) coupled with V2X simulation runtime infrastructure.

Development of Energy-aware Mobile Applications Based on Resource Outsourcing

Smart connected devices such as smartphones and tablets are battery-operated to facilitate their mobility. Therefore, low power consumption is a critical requirement for mobile hardware and for the software designed for such devices. In addition to efficient power management techniques and new battery technologies based on nanomaterials, cloud computing has emerged as a promising technique for reducing energy consumption as well as augmenting the computational and memory capabilities of mobile devices. In this study, we designed and implemented a framework that allows for the energy-efficient execution of mobile applications by partially offloading the workload of a mobile device onto a resourceful cloud. This framework comprises a development toolkit, which facilitates the development of mobile applications capable of supporting computation offloading, and a runtime infrastructure for deployment in the cloud. Using this framework, we implemented three different mobile applications and demonstrated that considerable energy savings can be achieved compared with local processing for both resource-intensive and lightweight applications, especially when using high-speed networks such as Wi-Fi and Long-Term Evolution.

A Kind of Distributed Simulation Runtime Infrastructure Based on Grid

Although high level architecture (HLA) is the mainstream of application in distributed simulation technology, poor interoperability, deficit reuse of the codes of the simulation applications and inferior expansibility of the simulation system had became the research focus of simulation community at home and broad. And Grid is able to manage a large number of heterogeneous resources of geographically and organizationally distributed. So a kind of Grid-based distribution simulation runtime infrastructure (GSRI) was put forward. Here with WSRF specification, GT4 tools was used to encapsulate the pRTI interfaces. And given the performance analysis, it is shown that despite the latency created by the transmission through wide area network (WAN), the GSRI may support the large-scare distributed simulation applications. It has a profound theoretical significance and great value of the engineering.

Distributed and Adaptive Business Process Execution

Centralized business process execution engines are not adequate to guarantee smooth process execution in the presence of multiple, concurrent, long-running process instances exchanging voluminous data. In the centralized architecture of most BPEL engine solutions, the execution of BPEL processes is performed in a closed runtime environment where process instances are isolated from each other, as well as from any other potential sources of information. This prevents processes from finding relative data at runtime to adapt their behavior in a dynamic manner. The goal of this chapter is to present a solution for the performance improvement of BPEL engines by using a distributed architecture that enables the scalable execution of service-oriented processes, while also supporting their data-driven adaptation. The authors propose a decentralized BPEL engine architecture using a hypercube peer-to-peer topology with data-driven adaptation capabilities that incorporates Artificial Intelligence (AI) planning and context-aware computing techniques to support the discovery of process execution paths at deployment time and improve the overall throughput of the execution infrastructure. The proposed solution is part of the runtime infrastructure that was developed for the environmental science industry to support the efficient execution and monitoring of service-oriented environmental science models.