We investigate the energy efficiency of a library designed for parallel computations with sparse matrices. The library leverages high-performance, energy-efficient Graphics Processing Unit (GPU) accelerators to enable large-scale scientific applications. Our primary development objective was to maximize parallel performance and scalability in solving sparse linear systems whose dimensions far exceed the memory capacity of a single node. To this end, we devised methods that expose a high degree of parallelism while optimizing algorithmic implementations for efficient multi-GPU usage. Previous work has already demonstrated the library’s performance efficiency on large-scale systems comprising thousands of NVIDIA GPUs, achieving improvements over state-of-the-art solutions. In this paper, we extend those results by providing energy profiles that address the growing sustainability requirements of modern HPC platforms. We present our methodology and tools for accurate runtime energy measurements of the library’s core components and discuss the findings. Our results confirm that optimizing GPU computations and minimizing data movement across memory and computing nodes reduces both time-to-solution and energy consumption. Moreover, we show that the library delivers substantial advantages over comparable software frameworks on standard benchmarks.

Linear solvers are key components in any software platform for scientific and engineering computing. The solution of large and sparse linear systems lies at the core of physics-driven numerical simulations relying on partial differential equations (PDEs) and often represents a significant bottleneck in data-driven procedures, such as scientific machine learning. In this paper, we present an efficient implementation of the preconditioned s-step Conjugate Gradient (CG) method, originally proposed by Chronopoulos and Gear in 1989, for large clusters of Nvidia GPU-accelerated computing nodes. The method, often referred to as communication-reduced or communication-avoiding CG, reduces global synchronizations and data communication steps compared to the standard approach, enhancing strong and weak scalability on parallel computers. Our main contribution is the design of a parallel solver that fully exploits the aggregation of low-granularity operations inherent to the s-step CG method to leverage the high throughput of GPU accelerators. Additionally, it applies overlap between data communication and computation in the multi-GPU sparse matrix-vector product. Experiments on classic benchmark datasets, derived from the discretization of the Poisson PDE, demonstrate the potential of the method.

ABSTRACT The paper focuses on the improvement of the existing nsparse Nagasaka et al. algorithm and its extension to the multi‐GPU setting for the application of real engineering problems. In this work, we propose a distributed multi‐GPU framework for SpGEMM that is designed specifically for the nsparse like algorithms. The results show ∼2 times speed‐up for nsparse and close to ideal scalability of the multi‐GPU extension with the number of GPUs. Finally, we test the proposed algorithm in the AMG setting by computing the double SpGEMM product.

The TEXTAROSSA project aims to bridge the technology gaps that exascale computing systems will face in the near future in order to overcome their performance and energy efficiency challenges. This project provides solutions for improved energy efficiency and thermal control, seamless integration of heterogeneous accelerators in HPC multi-node platforms, and new arithmetic methods. Challenges are tacked through a co-design approach to heterogeneous HPC solutions, supported by the integration and extension of HW and SW IPs, programming models, and tools derived from European research.

We present and release in open source format a sparse linear solver which efficiently exploits heterogeneous parallel computers. The solver can be easily integrated into scientific applications that need to solve large and sparse linear systems on modern parallel computers made of hybrid nodes hosting Nvidia Graphics Processing Unit (GPU) accelerators. The work extends previous efforts of some of the authors in the exploitation of a single GPU accelerator and proposes an implementation, based on the hybrid MPI-CUDA software environment, of a Krylov-type linear solver relying on an efficient Algebraic MultiGrid (AMG) preconditioner already available in theBootCMatchGlibrary. Our design for the hybrid implementation has been driven by the best practices for minimizing data communication overhead when multiple GPUs are employed, yet preserving the efficiency of the GPU kernels. Strong and weak scalability results of the new version of the library on well-known benchmark test cases are discussed. Comparisons with the Nvidia AmgX solution show a speedup, in the solve phase, up to 2.0x.

Abstract Tor is an open source software that allows accessing various kinds of resources, known as hidden services, while guaranteeing sender and receiver anonymity. Tor relies on a free, worldwide, overlay network, managed by volunteers, that works according to the principles of onion routing in which messages are encapsulated in layers of encryption, analogous to layers of an onion. The Tor Web is the set of web resources that exist on the Tor network, and Tor websites are part of the so-called dark web. Recent research works have evaluated Tor security, its evolution over time, and its thematic organization. Nevertheless, limited information is available about the structure of the graph defined by the network of Tor websites, not to be mistaken with the network of nodes that supports the onion routing. The limited number of entry points that can be used to crawl the network, makes the study of this graph far from being simple. In the present paper we analyze two graph representations of the Tor Web and the relationship between contents and structural features, considering three crawling datasets collected over a five-month time frame. Among other findings, we show that Tor consists of a tiny strongly connected component, in which link directories play a central role, and of a multitude of services that can (only) be reached from there. From this viewpoint, the graph appears inefficient. Nevertheless, if we only consider mutual connections, a more efficient subgraph emerges, that is, probably, the backbone of social interactions in Tor.

We investigate the security of split learning---a novel collaborative machine learning framework that enables peak performance by requiring minimal resource consumption. In the present paper, we expose vulnerabilities of the protocol and demonstrate its inherent insecurity by introducing general attack strategies targeting the reconstruction of clients' private training sets. More prominently, we show that a malicious server can actively hijack the learning process of the distributed model and bring it into an insecure state that enables inference attacks on clients' data. We implement different adaptations of the attack and test them on various datasets as well as within realistic threat scenarios. We demonstrate that our attack can overcome recently proposed defensive techniques aimed at enhancing the security of the split learning protocol. Finally, we also illustrate the protocol's insecurity against malicious clients by extending previously devised attacks for Federated Learning.

To achieve high performance and high energy efficiency on near-future exascale computing systems, three key technology gaps needs to be bridged. These gaps include: energy efficiency and thermal control; extreme computation efficiency via HW acceleration and new arithmetics; methods and tools for seamless integration of reconfigurable accelerators in heterogeneous HPC multi-node platforms. TEXTAROSSA aims at tackling this gap through a co-design approach to heterogeneous HPC solutions, supported by the integration and extension of HW and SW IPs, programming models and tools derived from European research.

Learning useful representations from unstructured data is one of the core challenges, as well as a driving force, of modern data-driven approaches. Deep learning has demonstrated the broad advantages of learning and harnessing such representations.In this paper, we introduce a deep generative model representation learning approach for password guessing. We show that an abstract password representation naturally offers compelling and versatile properties that open new directions in the extensively studied, and yet presently active, password guessing field. These properties can establish novel password generation techniques that are neither feasible nor practical with the existing probabilistic and non-probabilistic approaches. Based on these properties, we introduce: (1) A general framework for conditional password guessing that can generate passwords with arbitrary biases; and (2) an Expectation Maximization-inspired framework that can dynamically adapt the estimated password distribution to match the distribution of the attacked password set.

Summary GPUs are very powerful computing accelerators that are often employed in single‐device configuration. However, there is a steadily growing interest in using multiple GPUs in a concurrent way both to overcome the memory limitations of the single device and to further reduce execution times. Until recently, communication among GPUs had been carried out mainly by using networking technologies originally devised for standard CPUs with the CPU playing an active role in the communication. However, new alternatives start to be available in which a moderate number of GPUs are directly connected each other by means of proprietary technologies. We present the results of a set of experiments aimed at assessing the performance of some of these hardware/software platforms using a particularly challenging application as a benchmark. We release its source code to facilitate people interested in reproducing or extending our results.

We describe main issues and design principles of an efficient implementation, tailored to recent generations of Nvidia Graphics Processing Units (GPUs), of an Algebraic Multigrid (AMG) preconditioner previously proposed by one of the authors and already available in the open-source package BootCMatch: Bootstrap algebraic multigrid based on Compatible weighted Matching for standard CPU. The AMG method relies on a new approach for coarsening sparse symmetric positive definite (spd) matrices, named "coarsening based on compatible weighted matching". It exploits maximum weight matching in the adjacency graph of the sparse matrix, driven by the principle of compatible relaxation, providing a suitable aggregation of unknowns which goes beyond the limits of the usual heuristics applied in the current methods. We adopt an approximate solution of the maximum weight matching problem, based on a recently proposed parallel algorithm, referred as the Suitor algorithm, and show that it allow us to obtain good quality coarse matrices for our AMG on GPUs. We exploit inherent parallelism of modern GPUs in all the kernels involving sparse matrix computations both for the setup of the preconditioner and for its application in a Krylov solver, outperforming preconditioners available in Nvidia AmgX library. We report results about a large set of linear systems arising from discretization of scalar and vector partial differential equations (PDEs).

The exploration and analysis of Web graphs has flourished in the recent past, producing a large number of relevant and interesting research results. However, the unique characteristics of the Tor network demand for specific algorithms to explore and analyze it. Tor is an anonymity network that allows offering and accessing various Internet resources while guaranteeing a high degree of provider and user anonymity. So far the attention of the research community has focused on assessing the security of the Tor infrastructure. Most research work on the Tor network aimed at discovering protocol vulnerabilities to de-anonymize users and services, while little or no information is available about the topology of the Tor Web graph or the relationship between pages' content and topological structure. With our work we aim at addressing such lack of information. We describe the topology of the Tor Web graph measuring both global and local properties by means of well-known metrics that require due to the size of the network, high performance algorithms. We consider three different snapshots obtained by extensively crawling Tor three times over a 5 months time frame. Finally we present a correlation analysis of pages' semantics and topology, discussing novel insights about the Tor Web organization and its content. Our findings show that the Tor graph presents some of the characteristics of social and surface web graphs, along with a few unique peculiarities.

Tor hidden services allow offering and accessing various Internet resources while guaranteeing a high degree of provider and user anonymity. So far, most research work on the Tor network aimed at discovering protocol vulnerabilities to de-anonymize users and services. Other work aimed at estimating the number of available hidden services and classifying them. Something that still remains largely unknown is the structure of the graph defined by the network of Tor services. In this paper, we describe the topology of the Tor graph (aggregated at the hidden service level) measuring both global and local properties by means of well-known metrics. We consider three different snapshots obtained by extensively crawling Tor three times over a 5 months time frame. We separately study these three graphs and their shared “stable” core. In doing so, other than assessing the renowned volatility of Tor hidden services, we make it possible to distinguish time dependent and structural aspects of the Tor graph. Our findings show that, among other things, the graph of Tor hidden services presents some of the characteristics of social and surface web graphs, along with a few unique peculiarities, such as a very high percentage of nodes having no outbound links.

The exploration and analysis of Web graphs has flourished in the recent past, producing a large number of relevant and interesting research results. However, the unique characteristics of the Tor network limit the applicability of standard techniques and demand for specific algorithms to explore and analyze it. The attention of the research community has focused on assessing the security of the Tor infrastructure (i.e., its ability to actually provide the intended level of anonymity) and on discussing what Tor is currently being used for. Since there are no foolproof techniques for automatically discovering Tor hidden services, little or no information is available about the topology of the Tor Web graph. Even less is known on the relationship between content similarity and topological structure. The present article aims at addressing such lack of information. Among its contributions: a study on automatic Tor Web exploration/data collection approaches; the adoption of novel representative metrics for evaluating Tor data; a novel in-depth analysis of the hidden services graph; a rich correlation analysis of hidden services’ semantics and topology. Finally, a broad interesting set of novel insights/considerations over the Tor Web organization and content are provided.

We present the results obtained by using an evolution of our CUDA-based solution for the exploration, via a breadth first search, of large graphs. This latest version exploits at its best the features of the Kepler architecture and relies on a combination of techniques to reduce both the number of communications among the GPUs and the amount of exchanged data. The final result is a code that can visit more than 800 billion edges in a second by using a cluster equipped with 4,096 Tesla K20X GPUs.

We present simulations of blood and cancer cell separation in complex microfluidic channels with subcellular resolution, demonstrating unprecedented time to solution, performing at 65.5% of the available 39.4 PetaInstructions/s in the 18, 688 nodes of the Titan supercomputer.

We present performance results for the simulation of proteins suspensions in crowding conditions obtained with MUPHY, a computational platform for multi-scale simulations of real-life biofluidic problems. Previous versions of MUPHY have been used in the past for the simulation of blood flow through the human coronary arteries and DNA translocation across nanopores. The simulation exhibits excellent scalability up to 18, 000 K20X Nvidia GPUs and achieves almost 20 Petaflops of aggregate sustained performance with a peak performance of 27.5 Petaflops for the most intensive computing component. Those figures demonstrate once again the flexibility of MUPHY in simulating biofluidic phenomena, exploiting at their best the features of the architecture in use. Preliminary results were obtained in the present case on a completely different platform, the IBM Blue Gene/Q. The combination of novel mathematical models, computational algorithms, hardware technology, code tuning and parallelization techniques required to achieve these results are presented.

We present a computational framework for multi-scale simulations of real-life biofluidic problems. The framework allows to simulate suspensions composed by hundreds of millions of bodies interacting with each other and with a surrounding fluid in complex geometries. We apply the methodology to the simulation of blood flow through the human coronary arteries with a spatial resolution comparable with the size of red blood cells, and physiological levels of hematocrit (the red blood cell volume fraction). The simulation exhibits excellent scalability on a cluster of 4000 M2050 Nvidia GPUs and achieves close to 1 Petaflop aggregate performance, which demonstrates the capability to predicting the evolution of biofluidic phenomena of clinical significance. The combination of novel mathematical models, computational algorithms, hardware technology, code tuning and optimization required to achieve these results are presented.

We present the first large-scale simulation of blood flow in the coronary artieries and other vessels supplying blood to the heart muscle, with a realistic description of human arterial geometry at spatial resolutions from centimeters down to 10 microns (near the size of red blood cells). This multiscale simulation resolves the fluid into a billion volume units, embedded in a bounding space of 300 billion voxels, coupled with the concurrent motion of 300 million red blood cells, which interact with one another and with the surrounding fluid. The level of detail is sufficient to describe phenomena of potential physiological and clinical significance, such as the development of atherosclerotic plaques. The simulation achieves excellent scalability on up to 294, 912 Blue Gene/P computational cores.

Abstract We describe the porting of the Lattice Boltzmann component of MUPHY, a multi‐physics/scale simulation software, to multiple graphics processing units using the Compute Unified Device Architecture. The novelty of this work is the development ofad hoctechniques for optimizing the indirect addressing that MUPHY uses for efficient simulations of irregular domains. Copyright © 2009 John Wiley & Sons, Ltd.

Vaccine research is a combinatorial science requiring computational analysis of vaccine components, formulations and optimization. We have developed a framework that combines computational tools for the study of immune function and vaccine development. This framework, named ImmunoGrid combines conceptual models of the immune system, models of antigen processing and presentation, system-level models of the immune system, Grid computing, and database technology to facilitate discovery, formulation and optimization of vaccines. ImmunoGrid modules share common conceptual models and ontologies. The ImmunoGrid portal offers access to educational simulators where previously defined cases can be displayed, and to research simulators that allow the development of new, or tuning of existing, computational models. The portal is accessible at .

BACKGROUND: The efficiency of lymph nodes depends on tissue structure and organization, which allow the coordination of lymphocyte traffic. Despite their essential role, our understanding of lymph node specific mechanisms is still incomplete and currently a topic of intense research. RESULTS: In this paper, we present a hybrid discrete/continuous model of the lymph node, accounting for differences in cell velocity and chemotactic response, influenced by the spatial compartmentalization of the lymph node and the regulation of cells migration, encounter, and antigen presentation during the inflammation process. CONCLUSION: Our model reproduces the correct timing of an immune response, including the observed time delay between duplication of T helper cells and duplication of B cells in response to antigen exposure. Furthermore, we investigate the consequences of the absence of dendritic cells at different times during infection, and the dependence of system dynamics on the regulation of lymphocyte exit from lymph nodes. In both cases, the model predicts the emergence of an impaired immune response, i.e., the response is significantly reduced in magnitude. Dendritic cell removal is also shown to delay the response time with respect to normal conditions.

We present a parallel high-performance program that combines microscopic molecular dynamics (MD) with a mesoscopic Lattice Boltzmann (LB) method to simulate the translocation of biopolymers through nanometer size pores explicitly taking into account the interactions of the molecule with the surrounding fluid. The parallel implementation exhibits excellent scalability on the BlueGene platform and includes techniques which may improve the flexibility and efficiency of other complex multi-physics applications.

SockMi is a solution for the migration of TCP/IP connections between Linux systems. Only the migrating peer of the connection needs to reside on a Linux system. The migration is completely transparent to the other peer that can reside on a system running any operating system. Our solution does not require changes to existing Linux kernel data structures and algorithms and can be activated in any phase of the connection. Both 2.4 and 2.6 versions of the Linux kernel are supported.

MOTIVATION: Highly Active AntiRetroviral Therapies (HAART) can prolong life significantly to people infected by HIV since, although unable to eradicate the virus, they are quite effective in maintaining control of the infection. However, since HAART have several undesirable side effects, it is considered useful to suspend the therapy according to a suitable schedule of Structured Therapeutic Interruptions (STI). In the present article we describe an application of genetic algorithms (GA) aimed at finding the optimal schedule for a HAART simulated with an agent-based model (ABM) of the immune system that reproduces the most significant features of the response of an organism to the HIV-1 infection. RESULTS: The genetic algorithm helps in finding an optimal therapeutic schedule that maximizes immune restoration, minimizes the viral count and, through appropriate interruptions of the therapy, minimizes the dose of drug administered to the simulated patient. To validate the efficacy of the therapy that the genetic algorithm indicates as optimal, we ran simulations of opportunistic diseases and found that the selected therapy shows the best survival curve among the different simulated control groups. AVAILABILITY: A version of the C-ImmSim simulator is available at http://www.iac.cnr.it/~filippo/c-ImmSim.html

BACKGROUND: Protein interactions support cell organization and mediate its response to any specific stimulus. Recent technological advances have produced large data-sets that aim at describing the cell interactome. These data are usually presented as graphs where proteins (nodes) are linked by edges to their experimentally determined partners. This representation reveals that protein-protein interaction (PPI) networks, like other kinds of complex networks, are not randomly organized and display properties that are typical of "hierarchical" networks, combining modularity and local clustering to scale free topology. However informative, this representation is static and provides no clue about the dynamic nature of protein interactions inside the cell. RESULTS: To fill this methodological gap, we designed and implemented a computer model that captures the discrete and stochastic nature of protein interactions. In ProtNet, our simplified model, the intracellular space is mapped onto either a two-dimensional or a three-dimensional lattice with each lattice site having a linear size (5 nm) comparable to the diameter of an average globular protein. The protein filled lattice has an occupancy (e.g. 20%) compatible with the estimated crowding of proteins in the cell cytoplasm. Proteins or protein complexes are free to translate and rotate on the lattice that represents a sort of naïve unstructured cell (devoid of compartments). At each time step, molecular entities (proteins or complexes) that happen to be in neighboring cells may interact and form larger complexes or dissociate depending on the interaction rules defined in an experimental protein interaction network. This whole procedure can be seen as a sort of "discrete molecular dynamics" applied to interacting proteins in a cell. We have tested our model by performing different simulations using as interaction rules those derived from an experimental interactome of Saccharomyces cerevisiae (1378 nodes, 2491 edges) and we have compared the dynamics of complex formation in a two and a three dimensional lattice model. CONCLUSION: ProtNet is a cellular automaton model, where each protein molecule or complex is explicitly represented and where simple interaction rules are applied to populations of discrete particles. This tool can be used to simulate the dynamics of protein interactions in the cell.

The introduction of the IP multimedia subsystem on 3G cellular networks and the integration with other widely deployed wireless networks based on the IEEE 802.11 protocol family require support for both mobility and quality of service. When mobile systems move across heterogeneous networks, ongoing real-time sessions are affected not only by handoff delay but also by different packet delay and bit rate. In this paper, we propose a cross-layer mechanism that takes into account mobility at different layers of the network stack in order to yield better quality for VoIP, videoconferencing, and other real-time applications. We describe our cross-layer architecture, adaptation techniques, a prototype implementation, and experimental results.

We describe possible denial of service attacks to infrastructure wireless 802.11 networks. To carry out such attacks only commodity hardware and software components are required. The results show that serious vulnerabilities exist in different access points and that a single malicious station can easily hinder any legitimate communication within a basic service set.

We present a detailed analysis of the UNIX system calls and classify them according to their level of threat with respect to system penetration. Based on these results, an effective mechanism is proposed to control the invocation of critical, from the security viewpoint, system calls. The integration into existing UNIX operating systems is carried out by instrumenting the code of the system calls in such a way that the execution is granted only in the case where the invoking process and the value of the arguments comply with the rules held in an access control database. This method does not require changes in the kernel data structures and algorithms. All kernel modifications are transparent to the application processes that continue to work correctly with no need of source code changes or recompilation. A working prototype has been implemented as a loadable kernel module for the Linux operating system. The prototype is able to detect and block any attacks by which an intruder tries to gain direct access to the system as a privileged user.

We propose a cost-effective mechanism, to control the invocation of critical, from the security viewpoint, system calls. The integration into existing UNIX operating systems is carried out by instrumenting the code of the system calls so that the system call itself once invoked checks to see whether the invoking process and the argument values passed comply with the rules held in an access control database. This method provides simple interception of both system calls and their argument values and do not require changes in the kernel data structures and algorithms. All kernel modifications are transparent to the application processes that can continue to work correctly without needing changes of the source code or re-compilation. A working prototype has been implemented inside the kernel of the Linux operating system, the prototype is able to detect and block also buffer overflow based attacks.

Collective communication operations (CCOs) are one of the most powerful tools for parallel processing on distributed memory architectures. From the theoretical viewpoint there has been a major effort in the design of optimal algorithms for these operations, especially for massive parallel processors (MPPs). However, in spite of the increasing availability of MPPs, there are just a few limited experimental checks of the different theories, so the assessment of their real value is not easy. The aim of the present paper is to address such issues for the most common CCOs, considering practical algorithms that can be included in a generic communication library. The main result is a new algorithm for building a quasi-optimal broadcast tree that is much simpler than, and as efficient as, previously available algorithms. To investigate the advantages and drawbacks of the proposed algorithms, a large set of experimental data has been collected on an IBM SP2 parallel system. The data demonstrate the efficiency of our approach in a number of interesting cases. Finally, all the experimental results have been related to the model used in designing the algorithms. © 1998 John Wiley & Sons, Ltd.