publications

2024

1. Advanced Resource Allocation in the Context of Heterogeneous Workflows Management

   Lubrano Francesco, Vercellino Chiara, Vitali Giacomo,  Viviani Paolo, Scionti Alberto, and Terzo Olivier

   In Proceedings of the 2nd Workshop on Workflows in Distributed Environments Apr 2024

   Abs DOI

   In High-Performance Computing (HPC), workflows are utilized to define and manage a set of interdependent computations which allow the users to extract insights from (scientific) numerical simulations or data analytics. HPC platforms can perform extreme-scale simulations, combining Artificial Intelligence (AI) training and inference and data analytics (we refer to heterogeneous workflows), by providing tools and computing resources which serve a variety of use-cases spanning very diverse application domains (e.g., weather forecasting, quantum mechanics, etc.). Executing such workflows at scale requires to handle dependencies, job submission automation, I/O mechanisms. Despite State-of-the-Art batch schedulers can be configured and integrated with tools accomplishing this automation, a number of cases where resource allocation can lead to inefficiencies still exist. In this paper, to overcome these limitations, we present the WARP (Workflow-aware Advanced Resource Planner), a tool that integrates with workflow management tools and batch schedulers, to reserve in advance resources for an optimal execution of jobs, based on their duration, dependencies and machine load. WARP has been designed to minimize the overall workflow execution, without violating the priority policies for cluster users imposed by the system administrators.

2023

1. BBQ-mIS: A Parallel Quantum Algorithm for Graph Coloring Problems

   Vercellino Chiara, Vitali Giacomo,  Viviani Paolo, Giusto Edoardo, Scionti Alberto, Scarabosio Andrea, Terzo Olivier, and Montrucchio Bartolomeo

   In 2023 IEEE International Conference on Quantum Computing and Engineering (QCE) Sep 2023

   Abs DOI

   Among the limitations of current quantum machines, the qubits count represents one of the most critical challenges for porting reasonably large computational problems, such as those coming from real-world applications, to the scale of the quantum hardware. In this regard, one possibility is to decompose the problems at hand and exploit parallelism over multiple size-limited quantum resources. To this purpose, we designed a hybrid quantum-classical algorithm, i.e., BBQ-mIS, to solve graph coloring problems on Rydberg atoms quantum machines. The BBQ-mIS algorithm combines the natural representation of Maximum Independent Set (MIS) problems onto the machine Hamiltonian with a Branch&Bound (BB) approach to identify a proper graph coloring. In the proposed solution, the graph representation emerges from qubit interactions (qubits represent vertexes of the graph), and the coloring is then retrieved by iteratively assigning one color to a maximal set of independent vertexes of the graph, still minimizing the number of colors with the Branch&Bound approach. We emulated real quantum hardware onto an IBM Power9-based cluster, with 32 cores/node and 256 GB/node, and exploited an MPI-enhanced library to implement the parallelism for the BBQ-mIS algorithm. Considering this use case, we also identify some technical requirements and challenges for an effective HPC-QC integration. The results show that our problem decomposition is effective in terms of graph coloring solutions quality, and provide a reference for applying this methodology to other quantum technologies or applications.

2. Deep Learning for Real-Time Neural Decoding of Grasp

   Viviani Paolo, Gesmundo Ilaria, Ghinato Elios, Agudelo-Toro Andres, Vercellino Chiara, Vitali Giacomo, Bergamasco Letizia, Scionti Alberto, Ghislieri Marco, Agostini Valentina, Terzo Olivier, and Scherberger Hansjörg

   In Machine Learning and Knowledge Discovery in Databases: Applied Data Science and Demo Track Sep 2023

   Abs DOI

   Neural decoding involves correlating signals acquired from the brain to variables in the physical world like limb movement or robot control in Brain Machine Interfaces. In this context, this work starts from a specific pre-existing dataset of neural recordings from monkey motor cortex and presents a Deep Learning-based approach to the decoding of neural signals for grasp type classification. Specifically, we propose here an approach that exploits LSTM networks to classify time series containing neural data (i.e., spike trains) into classes representing the object being grasped.

3. Neural Optimization for Quantum Architectures: Graph Embedding Problems with Distance Encoder Networks

   Vercellino Chiara, Vitali Giacomo,  Viviani Paolo, Scionti Alberto, Scarabosio Andrea, Terzo Olivier, Giusto Edoardo, and Montrucchio Bartolomeo

   In 2023 IEEE 47th Annual Computers, Software, and Applications Conference (COMPSAC) Jun 2023

   Abs DOI url

   Quantum machines are among the most promising technologies expected to provide significant improvements in the following years. However, bridging the gap between real-world applications and their implementation on quantum hardware is still a complicated task. One of the main challenges is to represent through qubits (i.e., the basic units of quantum information) the problems of interest. According to the specific technology under-lying the quantum machine, it is necessary to implement a proper representation strategy, generally referred to as embedding. This paper introduces a neural-enhanced optimization framework to solve the constrained unit disk problem, which arises in the context of qubits positioning for neutral atoms-based quantum hardware. The proposed approach involves a modified autoencoder model, i.e., the Distances Encoder Network, and a custom loss, i.e., the Embedding Loss Function, respectively, to compute Euclidean distances and model the optimization constraints. The core idea behind this design relies on the capability of neural networks to approximate non-linear transformations to make the Distances Encoder Network learn the spatial transformation that maps initial non-feasible solutions of the constrained unit disk problem into feasible ones. The proposed approach outperforms classical solvers, given fixed comparable computation times, and paves the way to address other optimization problems through a similar strategy.

4. A Machine Learning Approach for an HPC Use Case: the Jobs Queuing Time Prediction

   Vercellino Chiara, Scionti Alberto, Varavallo Giuseppe,  Viviani Paolo, Vitali Giacomo, and Terzo Olivier

   Future Generation Computer Systems Jun 2023

   Abs DOI url

   High-Performance Computing (HPC) domain provided the necessary tools to support the scientific and industrial advancements we all have seen during the last decades. HPC is a broad domain targeting to provide both software and hardware solutions as well as envisioning methodologies that allow achieving goals of interest, such as system performance and energy efficiency. In this context, supercomputers have been the vehicle for developing and testing the most advanced technologies since their first appearance. Unlike cloud computing resources that are provided to the end-users in an on-demand fashion in the form of virtualized resources (i.e., virtual machines and containers), supercomputers’ resources are generally served through State-of-the-Art batch schedulers (e.g., SLURM, PBS, LSF, HTCondor). As such, the users submit their computational jobs to the system, which manages their execution with the support of queues. In this regard, predicting the behaviour of the jobs in the batch scheduler queues becomes worth it. Indeed, there are many cases where a deeper knowledge of the time experienced by a job in a queue (e.g., the submission of check-pointed jobs or the submission of jobs with execution dependencies) allows exploring more effective workflow orchestration policies. In this work, we focused on applying machine learning (ML) techniques to learn from the historical data collected from the queuing system of real supercomputers, aiming at predicting the time spent on a queue by a given job. Specifically, we applied both unsupervised learning (UL) and supervised learning (SL) techniques to define the most effective features for the prediction task and the actual prediction of the queue waiting time. For this purpose, two approaches have been explored: on one side, the prediction of ranges on jobs’ queuing times (classification approach) and, on the other side, the prediction of the waiting time at the minutes level (regression approach). Experimental results highlight the strong relationship between the SL models’ performances and the way the dataset is split. At the end of the prediction step, we present the uncertainty quantification approach, i.e., a tool to associate the predictions with reliability metrics, based on variance estimation.

2022

1. Accelerating Legacy Applications with Spatial Computing Devices

   Savio Paolo, Scionti Alberto, Vitali Giacomo,  Viviani Paolo, Vercellino Chiara, Terzo Olivier, Nguyen Huy-Nam, Magarielli Donato, Spano Ennio, Marconcini Michele, and Poli Francesco

   The Journal of Supercomputing Nov 2022

   Abs DOI

   Abstract Heterogeneous computing is the major driving factor in designing new energy-efficient high-performance computing systems. Despite the broad adoption of GPUs and other specialized architectures, the interest in spatial architectures like field-programmable gate arrays (FPGAs) has grown. While combining high performance, low power consumption and high adaptability constitute an advantage, these devices still suffer from a weak software ecosystem, which forces application developers to use tools requiring deep knowledge of the underlying system, often leaving legacy code (e.g., Fortran applications) unsupported. By realizing this, we describe a methodology for porting Fortran (legacy) code on modern FPGA architectures, with the target of preserving performance/power ratios. Aimed as an experience report, we considered an industrial computational fluid dynamics application to demonstrate that our methodology produces synthesizable OpenCL codes targeting Intel Arria10 and Stratix10 devices. Although performance gain is not far beyond that of the original CPU code (we obtained a relative speedup of \\\times\ \texttimes  0.59 and \\\times\ \texttimes  0.63, respectively, for a single optimized main kernel, while only on the Stratix10 we achieved \\\times\ \texttimes  2.56 by replicating the main optimized kernel 4 times), our results are quite encouraging to drawn the path for further investigations. This paper also reports some major criticalities in porting Fortran code on FPGA architectures.

2. Neural-Powered Unit Disk Graph Embedding: Qubits Connectivity for Some QUBO Problems

   Vercellino Chiara,  Viviani Paolo, Vitali Giacomo, Scionti Alberto, Scarabosio Andrea, Terzo Olivier, Giusto Edoardo, and Montrucchio Bartolomeo

   In 2022 IEEE International Conference on Quantum Computing and Engineering (QCE) Sep 2022

   Abs DOI url

   Graph embedding is a recurrent problem in quantum computing, for instance, quantum annealers need to solve a minor graph embedding in order to map a given Quadratic Unconstrained Binary Optimization (QUBO) problem onto their internal connectivity pattern. This work presents a novel approach to constrained unit disk graph embedding, which is encountered when trying to solve combinatorial optimization problems in QUBO form, using quantum hardware based on neutral Rydberg atoms. The qubits, physically represented by the atoms, are excited to the Rydberg state through laser pulses. Whenever qubits pairs are closer together than the blockade radius, entanglement can be reached, thus preventing entangled qubits to be simultaneously in the excited state. Hence, the blockade radius determines the adjacency pattern among qubits, corresponding to a unit disk configuration. Although it is straightforward to compute the adjacency pattern given the qubits’ coordinates, identifying a feasible unit disk arrangement that matches the desired QUBO matrix is, on the other hand, a much harder task. In the context of quantum optimization, this issue translates into the physical placement of the qubits in the 2D/3D register to match the machine’s Ising-like Hamiltonian with the QUBO formulation of the optimization problems. The proposed solution exploits the power of neural networks to transform an initial embedding configuration, which does not match the quantum hardware requirements or does not account for the unit disk property, into a feasible embedding properly representing the target optimization problems. Experimental results show that this new approach overcomes in performance Gurobi solver.

3. Taming Multi-node Accelerated Analytics: An Experience in Porting MATLAB to Scale with Python

   Viviani Paolo, Vitali Giacomo, Lengani Davide, Scionti Alberto, Vercellino Chiara, and Terzo Olivier

   In Complex, Intelligent and Software Intensive Systems Sep 2022

   DOI url
4. Dynamic Job Allocation on Federated Cloud-HPC Environments

   Vitali Giacomo, Scionti Alberto,  Viviani Paolo, Vercellino Chiara, and Terzo Olivier

   In Complex, Intelligent and Software Intensive Systems Sep 2022

   DOI url
5. Distributed HPC Resources Orchestration for Supporting Large-Scale Workflow Execution

   Scionti Alberto,  Viviani Paolo, Vitali Giacomo, Vercellino Chiara, Terzo Olivier, Hachinger Stephan, and Vojacek Lukáš

   In HPC, Big Data, and AI Convergence Towards Exascale: Challenge and Vision Jan 2022

   Abs DOI

   Artificial intelligence (AI) is gaining momentum in the scientific and industrial community for the ever-growing number of applications where such innovative techniques of learning form and processing large amount of data have proved successful. High-performance computing (HPC) and cloud resources providers are moving faster to be able to support new applications that benefit from the combination of traditional HPC simulation, machine learning and deep learning processing and big data analytics. However, the tighter the combination of these three elements is, the more complex the integration of innovative architectures into a single execution platform becomes. On one hand, application workflow management systems need to incorporate more functionalities and support dynamism in the execution, by preserving (energy) efficiency of the infrastructural resources. On the other hand, more exotic hardware accelerators (ranging from GPUs and FPGAs, to neural network processors (NNPs), to neuromorphic processors) need to be integrated in the computing assets in order to leverage performance boost. This chapter provides an overview of the future HPC, AI, and big-data cross-stack execution platform, as devised in the funded EuroHPC ACROSS project, which will be tailored to cope with all these challenges, and to support future exascale-ready applications.

6. Enabling the HPC and Artificial Intelligence Cross-Stack Convergence at the Exascale Level

   Scionti Alberto,  Viviani Paolo, Vitali Giacomo, Vercellino Chiara, and Terzo Olivier

   In HPC, Big Data, and AI Convergence Towards Exascale: Challenge and Vision Jan 2022

   Abs DOI

   Artificial intelligence (AI) is gaining momentum in the scientific and industrial community for the ever-growing number of applications where such innovative techniques of learning form and processing large amount of data have proved successful. High-performance computing (HPC) and cloud resources providers are moving faster to be able to support new applications that benefit from the combination of traditional HPC simulation, machine learning and deep learning processing and big data analytics. However, the tighter the combination of these three elements is, the more complex the integration of innovative architectures into a single execution platform becomes. On one hand, application workflow management systems need to incorporate more functionalities and support dynamism in the execution, by preserving (energy) efficiency of the infrastructural resources. On the other hand, more exotic hardware accelerators (ranging from GPUs and FPGAs, to neural network processors (NNPs), to neuromorphic processors) need to be integrated in the computing assets in order to leverage performance boost. This chapter provides an overview of the future HPC, AI, and big-data cross-stack execution platform, as devised in the funded EuroHPC ACROSS project, which will be tailored to cope with all these challenges, and to support future exascale-ready applications.

2021

1. HPC-Cloud-Big Data Convergent Architectures and Research Data Management: The LEXIS Approach

   Hachinger Stephan, Martinovič Jan, Terzo Olivier, Levrier Marc, Scionti Alberto, Magarielli Donato, Goubier Thierry, Parodi Antonio, Harsh Piyush, Apopei Florin-Ionut, Munke Johannes, García-Hernández Rubén, Golasowski Martin, Hayek Mohamad, Donnat Frédéric, Ganne Laurent, Koch-Hofer Cédric, Vitali Giacomo,  Viviani Paolo, Schorlemmer Danijel, Danovaro Emanuele, Parodi Andrea, Murphy Seán, and Dees Aaron

   In Proceedings of International Symposium on Grids & Clouds 2021 — PoS(ISGC2021) Jan 2021

   Abs DOI

   The LEXIS project (Large-scale EXecution for Industry & Society, H2020 GA825532) provides a platform for optimised execution of Cloud-HPC workflows, reducing computation time and increasing energy efficiency. The system will rely on advanced, distributed orchestration solutions (Atos YSTIA Suite, with Alien4Cloud and Yorc, based on TOSCA), the High-End Application Execution Middleware HEAppE, and new hardware capabilities for maximising efficiency in data processing, analysis and transfer (e.g. Burst Buffers with GPU- and FPGA-based data reprocessing). LEXIS handles computation tasks and data from three Pilots, based on representative and demanding HPC/Cloud-Computing use cases in Industry (SMEs) and Science: i) Simulations of complex turbomachinery and gearbox systems in Aeronautics, ii) Tsunami simulations and earthquake loss assessments which are time-constrained to enable immediate warnings and to support well-informed decisions, and iii) Weather and Climate simulations where massive amounts of in-situ data are assimilated to improve forecasts. A user-friendly LEXIS web portal, as a unique entry point, will provide access to data as well as workflow-handling and remote visualisation functionality. As part of its back-end, LEXIS builds an elaborate system for the handling of input, intermediate and result data. At its core, a Distributed Data Infrastructure (DDI) ensures the availability of LEXIS data at all participating HPC sites, which will be federated with a common LEXIS Authentication and Authorisation Infrastructure (with unified security model, user database and policies). The DDI leverages best of breed data-management solutions from EUDAT, such as B2SAFE (based on iRODS) and B2HANDLE. REST APIs on top of it will ensure a smooth interaction with LEXIS workflows and the orchestration layer. Last, but not least, the DDI will provide functionalities for Research Data Management following the FAIR principles (“Findable, Interoperable, Accessible, Reusable”), e.g. DOI acquisition, which helps to publish and disseminate open data products.

2. Towards Optimal Graph Coloring Using Rydberg Atoms

   Vitali Giacomo,  Viviani Paolo, Vercellino Chiara, Scarabosio Andrea, Scionti Alberto, Terzo Olivier, Giusto Edoardo, and Montrucchio Bartolomeo

   In The International Conference for High Performance Computing, Networking, Storage, and Analysis, Research posters Jan 2021

   Abs url

   Quantum mechanics is expected to revolutionize the computing landscape in the near future. Among the many candidate technologies for building universal quantum computers, Rydberg atoms-based systems stand out for being capable of performing both quantum simulations and working as gate-based universal quantum computers while operating at room temperature through an optical system. Moreover, they can potentially scale up to hundreds of quantum bits (qubits). In this work, we solve a Graph Coloring problem by iteratively computing the solutions of Maximal Independent Set (MIS) problems, exploiting the Rydberg blockade phenomenon. Experimental results using a simulation framework on the CINECA Marconi-100 supercomputer demonstrate the validity of the proposed approach.

2020

1. Authenticated and Auditable Data Sharing via Smart Contract

   Reniers Vincent, Gao Yuan, Zhang Ren,  Viviani Paolo, Madhusudan Akash, Lagaisse Bert, Nikova Svetla, Van Landuyt Dimitri, Lombardi Riccardo, Preneel Bart, and Joosen Wouter

   In Proceedings of the 35th Annual ACM Symposium on Applied Computing Jan 2020

   Abs DOI url

   Our main use case features multiple companies that iteratively optimize on the architectural properties of aircraft components in a decentralized manner. In each optimization step of the so-called multi-disciplinary optimization (MDO) process, sensitive data is exchanged between organizations, and we require auditability and traceability of actions taken to assure compliance with signed legal agreements.In this paper, we present a distributed protocol that coordinates authenticated and auditable exchanges of files, leveraging a smart contract. The entire life cycle of a file exchange, including file registration, access request and key distribution, is recorded and traceable via the smart contract. Moreover, when one party raises a dispute, the smart contract can be used to identify the dishonest party without compromising the file’s confidentiality.The proposed protocol provides a simple, novel, yet efficient approach to exchange files with support for data access auditability between companies involved in a private consortium with no incentive to share files outside of the protocol. We implemented the protocol in Solidity, deployed it on a private Ethereum blockchain, and validated it within the use case of a decentralized workflow.

2019

1. Deep Learning at Scale with Nearest Neighbours Communications

   Viviani Paolo

   Sep 2019

   Abs DOI url

   As deep learning techniques become more and more popular, there is the need to move these applications from the data scientist’s Jupyter notebook to efficient and reliable enterprise solutions. Moreover, distributed training of deep learning models will happen more and more outside the well-known borders of cloud and HPC infrastructure and will move to edge and mobile platforms. Current techniques for distributed deep learning have drawbacks in both these scenarios, limiting their long-term applicability. After a critical review of the established techniques for Data Parallel training from both a distributed computing and deep learning perspective, a novel approach based on nearest-neighbour communications is presented in order to overcome some of the issues related to mainstream approaches, such as global communication patterns. Moreover, in order to validate the proposed strategy, the Flexible Asynchronous Scalable Training (FAST) framework is introduced, which allows to apply the nearest-neighbours communications approach to a deep learning framework of choice. Finally, a relevant use-case is deployed on a medium-scale infrastructure to demonstrate both the framework and the methodology presented. Training convergence and scalability results are presented and discussed in comparison to a baseline defined by using state-of-the-art distributed training tools provided by a well-known deep learning framework.

2. Analysis of Architectural Variants for Auditable Blockchain-based Private Data Sharing

   Reniers Vincent, Van Landuyt Dimitri,  Viviani Paolo, Lagaisse Bert, Lombardi Riccardo, and Joosen Wouter

   In Proceedings of the 34th ACM/SIGAPP Symposium on Applied Computing Sep 2019

   Abs DOI url

   Many applications by design depend on costly trusted third-party auditors. One such example is the industrial application case of federated multi-disciplinary optimization (MDO), in which different organizations contribute to a complex engineering design effort. Although blockchain and distributed ledger technology (DLT) has strong potential in reducing the dependence on such intermediaries, the architectural complexity involved in designing a solution is daunting. In this paper, we analyze the architectural variants for decentralized private data sharing while guaranteeing auditability in terms of data access operations. Non-repudiation of actions taken by each party is a key requirement, as is availability of the shared data. % through storage governed by the chain. The architectural variants analyzed focus on attaining: (i) confidential data exchange, (ii) maintaining and governing access to the shared data, (iii) providing data access auditability, (iv) data validation or conflict resolution, and to a lesser degree (v) transaction and identity privacy. We systematically enumerate architectural decisions at the levels of: storage, policy-based file access control, data encryption methods, and auditability mechanisms for private data. This analysis is based on extensive assessment of the state of the art on decentralized private data access management using static or dynamic policies, and private data validation without exposing confidential information. The main contribution of this work is a comprehensive overview of architectural variants for decentralized control of private, encrypted data, and the involved trade-offs in terms of performance, auditable trust and security. These findings are validated in the context on the aforementioned industry case that involves federated multi-disciplinary optimization (MDO).

3. Accelerating spectral graph analysis through wavefronts of linear algebra operations

   Drocco Maurizio,  Viviani Paolo, Colonnelli Iacopo, Aldinucci Marco, and Grangetto Marco

   In Proc. of 27th Euromicro Intl. Conference on Parallel Distributed and network-based Processing (PDP) Sep 2019

   Abs DOI url

   The wavefront pattern captures the unfolding of a parallel computation in which data elements are laid out as a logical multidimensional grid and the dependency graph favours a diagonal sweep across the grid. In the emerging area of spectral graph analysis, the computing often consists in a wavefront running over a tiled matrix, involving expensive linear algebra kernels. While these applications might benefit from parallel heterogeneous platforms (multi-core with GPUs),programming wavefront applications directly with high-performance linear algebra libraries yields code that is complex to write and optimize for the specific application. We advocate a methodology based on two abstractions (linear algebra and parallel pattern-based run-time), that allows to develop portable, self-configuring, and easy-to-profile code on hybrid platforms.

4. Deep Learning at Scale

   Viviani Paolo, Drocco Maurizio, Baccega Daniele, Colonnelli Iacopo, and Aldinucci Marco

   In Proc. of 27th Euromicro Intl. Conference on Parallel Distributed and network-based Processing (PDP) Sep 2019

   Abs DOI url

   This work presents a novel approach to distributed training of deep neural networks (DNNs) that aims to overcome the issues related to mainstream approaches to data parallel training. Established techniques for data parallel training are discussed from both a parallel computing and deep learning perspective, then a different approach is presented that is meant to allow DNN training to scale while retaining good convergence properties. Moreover, an experimental implementation is presented as well as some preliminary results.

2018

1. Pushing the boundaries of parallel Deep Learning - A practical approach

   Viviani Paolo, Drocco Maurizio, and Aldinucci Marco

   CoRR Sep 2018
2. HPC4AI, an AI-on-demand federated platform endeavour

   Aldinucci Marco, Rabellino Sergio, Pironti Marco, Spiga Filippo,  Viviani Paolo, Drocco Maurizio, Guerzoni Marco, Boella Guido, Mellia Marco, Margara Paolo, Drago Idillio, Marturano Roberto, Marchetto Guido, Piccolo Elio, Bagnasco Stefano, Lusso Stefano, Vallero Sara, Attardi Giuseppe, Barchiesi Alex, Colla Alberto, and Galeazzi Fulvio

   In ACM Computing Frontiers May 2018

   Abs DOI url

   In April 2018, under the auspices of the POR-FESR 2014-2020 program of Italian Piedmont Region, the Turin’s Centre on High-Performance Computing for Artificial Intelligence (HPC4AI) was funded with a capital investment of 4.5Me and it began its deployment. HPC4AI aims to facilitate scientific research and engineering in the areas of Artificial Intelligence and Big Data Analytics. HPC4AI will specifically focus on methods for the on-demand provisioning of AI and BDA Cloud services to the regional and national industrial community, which includes the large regional ecosystem of Small-Medium Enterprises (SMEs) active in many different sectors such as automotive, aerospace, mechatronics, manufacturing, health and agrifood.

3. Scaling Dense Linear Algebra on Multicore and Beyond: a Survey

   Viviani Paolo, Drocco Maurizio, and Aldinucci Marco

   In Proc. of 26th Euromicro Intl. Conference on Parallel Distributed and network-based Processing (PDP) May 2018

   Abs DOI url

   The present trend in big-data analytics is to exploit algorithms with (sub-)linear time complexity, in this sense it is usually worth to investigate if the available techniques can be approximated to reach an affordable complexity. However, there are still problems in data science and engineering that involve algorithms with higher time complexity, like matrix inversion or Singular Value Decomposition (SVD). This work presents the results of a survey that reviews a number of tools meant to perform dense linear algebra at “Big Data” scale: namely, the proposed approach aims first to define a feasibility boundary for the problem size of shared-memory matrix factorizations, then to understand whether it is convenient to employ specific tools meant to scale out such dense linear algebra tasks on distributed platforms. The survey will eventually discuss the presented tools from the point of view of domain experts (data scientist, engineers), hence focusing on the trade-off between usability and performance.

4. Scientific Workflows on Clouds with Heterogeneous and Preemptible Instances

   Tordini Fabio, Aldinucci Marco,  Viviani Paolo, Merelli Ivan, and Liò Pietro

   In Proc. of the Intl. Conference on Parallel Computing, ParCo 2017, 12-15 September 2017, Bologna, Italy May 2018

   Abs DOI url

   The cloud environment is increasingly appealing for the HPC community, which has always dealt with scientific applications. However, there is still some skepticism about moving from traditional physical infrastructures to virtual HPC clusters. This mistrusting probably originates from some well known factors, including the effective economy of using cloud services, data and software availability, and the longstanding matter of data stewardship. In this work we discuss the design of a framework (based on Mesos) aimed at achieving a cost-effective and efficient usage of heterogeneous Processing Elements (PEs) for workflow execution, which supports hybrid cloud bursting over preemptible cloud Virtual Machines.

5. A Flexible Numerical Framework for Engineering—A Response Surface Modelling Application

   Viviani P., Aldinucci M., d’Ippolito R., Lemeire J., and Vucinic D.

   May 2018

   Abs DOI url

   This work presents an innovative approach adopted for the development of a new numerical software framework for accelerating dense linear algebra calculations and its application within an engineering context. In particular, response surface models (RSM) are a key tool to reduce the computational effort involved in engineering design processes like design optimization. However, RSMs may prove to be too expensive to be computed when the dimensionality of the system and/or the size of the dataset to be synthesized is significantly high or when a large number of different response surfaces has to be calculated in order to improve the overall accuracy (e.g. like when using ensemble modelling techniques). On the other hand, the potential of modern hybrid hardware (e.g. multicore, GPUs) is not exploited by current engineering tools, while they can lead to a significant performance improvement. To fill this gap, a software framework is being developed that enables the hybrid and scalable acceleration of the linear algebra core for engineering applications and especially of RSMs calculations with a user-friendly syntax that allows good portability between different hardware architectures, with no need of specific expertise in parallel programming and accelerator technology. The effectiveness of this framework is shown by comparing an accelerated code to a single-core calculation of a radial basis function RSM on some benchmark datasets. This approach is then validated within a real-life engineering application and the achievements are presented and discussed.

2017

1. Multiple Back-End Support for the Armadillo Linear Algebra Interface

   Viviani Paolo, Aldinucci Marco, Torquati Massimo, and d’lppolito Roberto

   In Proceedings of the Symposium on Applied Computing May 2017

   Abs DOI url

   The Armadillo C++ library provides programmers with a high-level Matlab-like syntax for linear algebra. Its design aims at providing a good balance between speed and ease of use. It can be linked with different back-ends, i.e. different LAPACK-compliant libraries. In this work we present a novel run-time support of Armadillo, which gracefully extends mainstream implementation to enable back-end switching without recompilation and multiple back-end support. The extension is specifically designed to not affect Armadillo class template prototypes, thus to be easily interoperable with future evolutions of the Armadillo library itself. The proposed software stack is then tested for functionality and performance against a kernel code extracted from an industrial application.

2016

1. An hybrid linear algebra framework for engineering

   Viviani Paolo, Aldinucci Marco, and d’Ippolito Roberto

   In Advanced Computer Architecture and Compilation for High-Performance and Embedded Systems (ACACES) – Poster Abstracts Jul 2016

   Abs url

   The aim of this work is to provide developers and domain experts with simple (Matlab-like) inter- face for performing linear algebra tasks while retaining state-of-the-art computational speed. To achieve this goal we extend Armadillo C++ library is extended in order to support with multiple LAPACK-compliant back-ends targeting different architectures including CUDA GPUs; moreover our approach involves the possibility of dynamically switching between such back-ends in order to select the one which is most convenient based on the specific problem and hardware configura- tion. This approach is eventually validated within an industrial environment.

2. A flexible numerical framework for engineering - a Response Surface Modelling application

   Viviani Paolo, Aldinucci Marco, d’Ippolito Roberto, Lemeire Jean, and Vucinic Dean

   In 10th Intl. Conference on Advanced Computational Engineering and Experimenting (ACE-X) Jul 2016

   Abs

   This work presents the innovative approach adopted for the development of a new numerical software framework for accelerating Dense Linear Algebra calculations and its application within an engineering context. In particular, Response Surface Models (RSM) are a key tool to reduce the computational effort involved in engineering design processes like design optimization. However, RSMs may prove to be too expensive to be computed when the dimensionality of the system and/or the size of the dataset to be synthesized is significantly high or when a large number of different Response Surfaces has to be calculated in order to improve the overall accuracy (e.g. like when using Ensemble Modelling techniques). On the other hand, it is a known challenge that the potential of modern hybrid hardware (e.g. multicore, GPUs) is not exploited by current engineering tools, while they can lead to a significant performance improvement. To fill this gap, a software framework is being developed that enables the hybrid and scalable acceleration of the linear algebra core for engineering applications and especially of RSMs calculations with a user-friendly syntax that allows good portability between different hardware architectures, with no need of specific expertise in parallel programming and accelerator technology. The effectiveness of this framework is shown by comparing an accelerated code to a single-core calculation of a Radial Basis Function RSM on some benchmark datasets. This approach is then validated within a real-life engineering application and the achievements are presented and discussed.

2015

1. Parallel Computing Techniques for High Energy Physics

   Viviani Paolo

   Jul 2015

   Abs

   Modern experimental achievements, with LHC results as a prominent but not exclusive representative, have undisclosed a new range of challenges concerning theoretical com- putations. Tree level QED calculation are no more satisfactory due to the very small experimental uncertainty of precision e+ e- measurements, so Next To Leading and Next to Next to Leading Order calculations are required. At the same time many-legs, high-order QCD processes needed to simulate LHC events are raising even more the bar of computational complexity. The drive for the present work has been the interest in calculating high multiplicity Higgs boson processes with a dedicated software library (RECOLA) currently under development at the University of Torino, as well as the related technological challenges. This thesis undertakes the task of exploring the possibilities offered by present and upcoming computing technologies in order to face these challenges properly. The first two chapters outlines the theoretical context and the available technologies. In chapter 3 a a case study is examined in full detail, in order to explore the suitability of different parallel computing solutions. In the chapter 4, some of those solutions are implemented in the context of the RECOLA library, allowing it to handle processes at a previously unexplored scale of complexity. Alongside, the potential of new, cost-effective parallel architectures is tested.