Search

Proceedings of: Third International Workshop on Sustainable Ultrascale Computing Systems (NESUS 2016). Sofia (Bulgaria), October, 6-7, 2016. As we move towards ultrascale computing, computer architecture is bound to see dramatic changes. Multiple nodes, with or without shared memory, multicore and accelerators (GPUs, FPGAs) will be the norm. For many problems, such as finite difference numerical simulations, the array used to represent a perfect match between the user level code and the hardware architecture’s uniform memory access. Arrays, and to some extent multiarrays, are well supported by most programming languages. A standard compiler maps the array for uniform memory. Some programming models, such as partitioned global address space, allows mapping an array across distributed, yet for each partition, uniform memory. For ultrascale architectures, the simple mapping between user level (multi)array and distributed, non-uniform memory, will disappear. Here we propose an API for arrays, empowering the software developer to implement their own array-memory layout. Application code written towards the API will be independent of underlying architecture changes, thus easily ported to new architectures as they evolve. This research has in part been financed by The Research Council of Norway through the project Design of a Mouldable Programming Language (DMPL), and has received support from EU under the COST Program Action IC1305, Network for Sustainable Ultrascale Computing (NESUS).

Proceedings of: Third International Workshop on Sustainable Ultrascale Computing Systems (NESUS 2016). Sofia (Bulgaria), October, 6-7, 2016. The modern high-performance computing systems (HPCS) are composed of hundreds of thousand computational nodes. An effective resource allocation in HPCS is a subject for many scientific research investigations. Many programming models for effective resources allocation have been proposed. The main purpose of those models is to increase the parallel performance of the HPCS. This paper investigates the efficiency of parallel algorithm for resource management optimization based on Artificial Bee Colony (ABC) metaheuristic while solving a package of NP-complete problems on multi-processor platform.In order to achieve minimal parallelization overhead in each cluster node, a multi-level hybrid programming model is proposed that combines coarse-grain and fine-grain parallelism. Coarse-grain parallelism is achieved through domain decomposition by message passing among computational nodes using Message Passing Interface (MPI) and fine-grain parallelism is obtained by loop-level parallelism inside each computation node by compiler-based thread parallelization via Intel TBB. Parallel communications profiling is made and parallel performance parameters are evaluated on the basis of experimental results.

Proceedings of: Third International Workshop on Sustainable Ultrascale Computing Systems (NESUS 2016). Sofia (Bulgaria), October, 6-7, 2016. The main goal of this paper is the suitability assessment of the OpenMP Accelerator Model (OMPAM) for porting a real-life scientific application to heterogeneous platforms containing a single Intel Xeon Phi coprocessor. This OpenMP extension is supported from version 4.0 of the standard, offering an unified directive-based programming model dedicated for massively parallel accelerators. In our study, we focus on applying the OMPAM extension together with the OpenMP tasks for a parallel application which implements the numerical model of alloy solidification. To map the application efficiently on target hybrid platforms using such constructs as omp target, omp target data and omp target update, we propose a decomposition of main tasks belonging to the computational core of the studied application. In consequence, the coprocessor is used to execute the major parallel workloads, while CPUs are responsible for executing a part of the application that do not require massively parallel resources. Effective overlapping computations with data transfers is another goal achieved in this way. The proposed approach allows us to execute the whole application 3.5 times faster than the original parallel version running on two CPUs. This research was conducted with the support of COST Action IC1305 (NESUS), as well as the National Science Centre (Poland) under grant no. UMO-2011/03/B/ST6/03500. The authors are grateful to the Czestochowa University of Technology for granting access to Intel Xeon Phi coprocessors provided by the MICLAB project no. POIG.02.03.00.24-093/13 (http://miclab.pl).

Proceedings of: Third International Workshop on Sustainable Ultrascale Computing Systems (NESUS 2016). Sofia (Bulgaria), October, 6-7, 2016. The power consumption of data centers is becoming a crucial challenge in the context of the steadily increasing demand for computation. In this regard finding a way to improve energy efficiency of running applications in data centers is becoming a crucial trend. One method to improve the processor utilization is the consolidation of applications on physical servers. It is possible to run multiple jobs in parallel on the same machine, especially when their requirements regarding computation are smaller than the maximum processor performance. It reduces the number of servers in the data center required to handle multiple requests and therefore leads to energy usage reductions. In this paper, we introduce a realistic model of applications with deadlines executed in parallel on a server and competing for the shared resources and present an energy-aware algorithm which may be used to minimize the overall energy consumption of the servers. European Community's Seventh Framework Program This work is partially supported by EU under the COST Program Action 1305: Network for Sustainable Ultrascale Computing (NESUS). The research presented in this paper is partially funded by a grant from Polish National Science Center under award number 2013/08/A/ST6/00296. This research was supported by the EU Seventh Framework Programme FP7/2007–2013 under grant agreement no. FP7-ICT-2013-10 (609757).

Proceedings of: Third International Workshop on Sustainable Ultrascale Computing Systems (NESUS 2016). Sofia (Bulgaria), October, 6-7, 2016. In this paper we address the issue of managing different energy sources which supply green powered datacenters. The sources are scheduled based on a priority scheme, aiming to maximize the renewable energy utilization, minimize the energy used from the grid and optimize battery usage. Dynamic power capping technique is used to put a threshold on the drawn energy from the grid. The algorithm is implemented and tested in CloudSim simulator. Renewable energy is considered as solar energy. A workload scheduling algorithm is already implemented for higher renewable energy utilization. The results show that the proposed scheme is efficient and it is a promising direction in the field of the optimization in datacenters using renewable energy. The work presented in this paper has been supported by EU under the COST programme Action IC1305: Network for Sustainable Ultrascale Computing (NESUS).

Proceedings of: Third International Workshop on Sustainable Ultrascale Computing Systems (NESUS 2016). Sofia (Bulgaria), October, 6-7, 2016. As data-intensive scientific prevalence arises, there is a necessity of simplifying the development, deployment, and execution of complex data analysis applications. The Data Mining Cloud Framework is a service-oriented system for allowing users to design and execute data analysis applications, defined as workflows, on cloud platforms, relying on cloud-provided storage services for I/O operations. Hercules is an in-memory I/O solution that can be deployed as an alternative to cloud storage services, providing additional performance and flexibility features. This work extends the DMCF-Hercules cooperation by applying novel data placement and task scheduling techniques for exposing and exploiting data locality in data-intensive workflows. This work is partially supported by EU under the COST Program Action IC1305: Network for Sustainable Ultrascale Computing (NESUS).

Proceedings of: Third International Workshop on Sustainable Ultrascale Computing Systems (NESUS 2016). Sofia (Bulgaria), October, 6-7, 2016. The speedup that can be achieved with parallel and distributed architectures is limited at least by two laws: the Amdahl’s and Gustafson’s laws. The former limits the speedup to a constant value when a fixed size problem is executed on a multiprocessor, while the latter limits the speedup up to its linear value for the fixed time problems, which means that it is limited by the number of used processors. However, a superlinear speedup can be achieved (speedup greater than the number of used processors) due to insufficient memory, while, parallel and, especially distributed systems can even slowdown the execution due to the communication overhead, when compared to the sequential one. Since the cloud performance is uncertain and it can be influenced by available memory and networks, in this paper we investigate if it follows the same speedup pattern as the other traditional distributed systems. The focus is to determine how the elastic cloud services behave in the different scaled environments. We define several scaled systems and we model the corresponding performance indicators. The analysis shows that both laws limit the speedup for a specific range of the input parameters and type of scaling. Even more, the speedup in cloud systems follows the Gustafson’s extreme cases, i.e. insufficient memory and communication bound domains. This work is partially supported by the European Union’s Horizon 2020 research and innovation programme under the grant agreements 644179 ENTICE: dEcentralized repositories for traNsparent and efficienT vIrtual maChine opErations (first two authors) and 643946, CloudLightning: Selforganizing, self-managing Heterogeneous Clouds (fourth author). The authors would like to acknowledge networking support by the COST programme Action IC1305, Network for Sustainable Ultrascale Computing (NESUS). European Commission

Proceedings of: Third International Workshop on Sustainable Ultrascale Computing Systems (NESUS 2016). Sofia (Bulgaria), October, 6-7, 2016. Ultrascale computing is required for many important applications in chemistry, computational fluid dynamics etc., see an overview in the paper Applications for Ultrascale Computing by M. Mihajlovic et al. published in the International Journal Supercomputing Frontiers and Innovations, Vol 2 (2015). In this abstract we shortly describe an application that involves many aspects described in the above paper - the multiscale material design problem. The problem of interest is analysis of the fiber reinforced concrete and we focus on modelling of stiffness through numerical homogenization and computing local material properties by inverse analysis. Both problems require a repeated solution of large-scale finite element problems up to 200 million degrees of freedom and therefore the importance of HPC and ultrascale computing is evident. The work is supported by COST Action IC1305 project Network for Sustainable Ultrascale Computing and a bilateral project of collaboration between the Institute of Geonics CAS and IICT BAS. Further support is through the projects LD15105 Ultrascale computing in geosciences and LQ1602 IT4Innovations excellence in science supported by the Ministry of Education, Youth and Sports of the Czech Republic.

Proceedings of: Third International Workshop on Sustainable Ultrascale Computing Systems (NESUS 2016). Sofia (Bulgaria), October, 6-7, 2016. Nek5000 is an open-source code for simulating incompressible flows using MPI for parallel communication. In the Nek5000 code, the tensor-product-based operator evaluation can be implemented as small dense matrix-matrix multiplications. It is clear that the routines for calculating the matrix-matrix product dominate the execution time of Nek5000. In this paper, we conduct the optimization of matrix-matrix multiplication using SIMD intrinsics and the LIBXSMM package. The evaluation of the computational cost and optimization of these subroutines is not only applied to the CFD code Nek5000, but also to the NekCEM and NekLEM software, which share same data structures with Nek5000.

Proceedings of: Third International Workshop on Sustainable Ultrascale Computing Systems (NESUS 2016). Sofia (Bulgaria), October, 6-7, 2016. The ever growing complexity of high performance computing systems imposes significant challenges to exploit as much as possible their computational and memory resources. Recently, the Cache-aware Roofline Model has gained popularity due to its simplicity when modeling multi-cores with complex memory hierarchy, characterizing applications bottlenecks, and quantifying achieved or remaining improvements. In this short paper we involve hardware locality topology detection to build the Cache Aware Roofline Model for modern processors in an open-source locality-aware tool. The proposed tool also includes a set of specific micro-benchmarks to assess the micro-architecture performance upper-bounds. The experimental results show that by relying on the proposed tool, it was possible to reach near-theoretical bounds of an Intel 3770K processor, thus proving the effectiveness of the modeling methodology. We would like to acknowledge Action IC1305 (NESUS) for funding this work.

Proceedings of: Third International Workshop on Sustainable Ultrascale Computing Systems (NESUS 2016). Sofia (Bulgaria), October, 6-7, 2016. The work presented here is an experimental study of performance in execution time and energy consumption of matrix multiplications on a heterogeneous server. The server features three different devices: a multicore CPU, an NVIDIA Tesla GPU, and an Intel Xeon Phi coprocessor. Matrix multiplication is one of the most used linear algebra kernels and, consequently, applications that make an intensive use of this operation can greatly benefit from efficient implementations. This is the case of the evaluation of matrix polynomials, a core operation used to calculate many matrix functions, which involve a very large number of products of square matrices. Although there exist many proposals for efficient implementations of matrix multiplications in heterogeneous environments, it is still difficult to find packages providing a matrix multiplication routine that is so easy to use, efficient, and versatile as its homogeneous counterparts. Our approach here is based on a simple implementation using OpenMP sections. We have also devised a functional model for the execution time that has been successfully applied to the evaluation of matrix polynomials of large degree so that it allows to balance the workload and minimizes the runtime cost.

Proceedings of: Third International Workshop on Sustainable Ultrascale Computing Systems (NESUS 2016). Sofia (Bulgaria), October, 6-7, 2016. In this work, we consider a parallel numerical solution of problems depending onf ractional power sof elliptic operator. Three different state of theart approaches are used to transform the original non-local problem into well-known local PDE problems. Parallel numerical algorithms for allt hreea pproaches are developed and discussed. Results of their parallel performance tests are presented and analysed. The work presented in this paper has been partially supported by EU under the COST programme Action IC1305,’Network for Sustainable Ultrascale Computing (NESUS)’. This research was also funded by agrant (No. MIP-074/2015) from the Research Council of Lithuania.

Proceedings of: Second International Workshop on Sustainable Ultrascale Computing Systems (NESUS 2015). Krakow (Poland), September 10-11, 2015. The work has been partially supported by EU under the COST Action IC1305 “Network for Sustainable Ultrascale Computing (NESUS)”.

Proceedings of: Second International Workshop on Sustainable Ultrascale Computing Systems (NESUS 2015). Krakow (Poland), September 10-11, 2015. High-Performance Computing has become an essential tool in numerous natural sciences. The modern highperformance computing systems are composed of hundreds of thousands of computational nodes, as well as deep memory hierarchies and complex interconnect topologies. Existing high performance algorithms and tools already require courageous programming and optimization efforts to achieve high efficiency on current supercomputers. On the other hand, these efforts are platform-specific and non-portable. A core challenge while solving NP-complete problems is the need to process these data with highly effective algorithms and tools where the computational costs grow exponentially. This paper investigates the efficiency of Nature-Inspired optimization algorithm for solving NP-complete problems, based on Artificial Bee Colony (ABC) metaheuristic. Parallel version of the algorithm have been proposed based on the flat parallel programming model with message passing for communication between the computational nodes in the platform and parallel programming model with multithreading for communication between the cores inside the computational node. Parallel communications profiling is made and parallel performance parameters are evaluated on the basis of experimental results. The results reported in this paper are part of the research project, Center of excellence "Supercomputing Applications" - DCVP 02/1, supported by the National Science Fund, Bulgarian Ministry of Education and Science.

Proceedings of: Second International Workshop on Sustainable Ultrascale Computing Systems (NESUS 2015). Krakow (Poland), September 10-11, 2015. Optimizing lightweight structures with numerical simulations leads to the development of complex simulation codes with high computational demands. The optimization approach for lightweight structures consisting of fiberreinforced plastics is considered. During the simulated optimization, independent simulation tasks have to be executed efficiently on the heterogeneous computing resources. In this article, several scheduling methods for distributing parallel simulation tasks among compute nodes are presented. Performance results are shown for the scheduling and execution of synthetic benchmark tasks, matrix multiplication tasks, as well as FEM simulation tasks on a heterogeneous compute cluster. This work was performed within the Federal Cluster of Excellence EXC 1075 “MERGE Technologies for Multifunctional Lightweight Structures” and supported by the German Research Foundation (DFG).

Proceedings of: Second International Workshop on Sustainable Ultrascale Computing Systems (NESUS 2015). Krakow (Poland), September 10-11, 2015. Building Ultrascale computer systems is a hard problem, not yet solved and fully explored. Combining the computing resources of multiple organizations, often in different administrative domains with heterogeneous hardware and diverse demands on the system, requires new tools and frameworks to be put in place. During previous work we developed POP-Java, a Java programming language extension that allows to easily develop distributed applications in a heterogeneous environment. We now present an extension to the POP-Java language, that allows to create application centered networks in which any member can benefit from the computing power and storage capacity of its members. An accounting system is integrated, allowing the different members of the network to bill the usage of their resources to the other members, if so desired. The system is expanded through a similar process as seen in social networks, making it possible to use the resources of friend and friends of friends. Parts of the proposed system has been implemented as a prototype inside the POP-Java programming language.

Proceedings of: Second International Workshop on Sustainable Ultrascale Computing Systems (NESUS 2015). Krakow (Poland), September 10-11, 2015. The recent activities to construct exascale and ultrascale distributed computational systems are opening a possibility to apply parallel and distributed computing techniques for applied problems which previously were considered as not solvable with the standard computational resources. In this paper we consider one global optimization problem where a set of feasible solutions is discrete and very large. There is no possibility to apply some apriori estimation techniques to exclude an essential part of these elements from the computational analysis, e.g. applying branch and bound type methods. Thus a full search is required in order to solve such global optimization problems. The considered problem describes visual cryptography algorithms. The main goal is to find optimal perfect gratings, which can guarantee high quality and security of the visual cryptography method. The full search parallel algorithm is based on master-slave paradigm. We present a library of C++ templates that allow the developer to implement parallel master-slave algorithms for his application without any parallel programming and knowledge of parallel programming API. These templates automatically give parallel solvers tailored for clusters of computers using MPI API and distributed computing applications using BOINC API. Results of some computational experiments are presented. The work presented in this paper has been partially supported by EU under the COST programme Action IC1305, ’Network for Sustainable Ultrascale Computing (NESUS)’.

Proceedings of: Second International Workshop on Sustainable Ultrascale Computing Systems (NESUS 2015). Krakow (Poland), September 10-11, 2015. IaaS cloud simulators are frequently used for evaluating new scheduling practices. Unfortunately, most of these simulators scarcely allow the evaluation of larger-scale cloud infrastructures (i.e., with physical machine counts over a few thousand). Thus, they are seldom applicable for evaluating infrastructures available in commercial cloud settings (e.g., users mostly do not wait for simulations to complete in such settings). DISSECT-CF was shown to be better scaling than several other simulators, but peta-scale infrastructures with often millions of CPU cores were out of scope for DISSECT-CF as well. This paper reveals a hierarchical scheduler extension of DISSECT-CF that not only allows its users to evaluate peta-scale infrastructure behaviour, but also opens possibilities for analysing new multi-cloud scheduling techniques. The paper then analyses the performance of the extended simulator through large-scale synthetic workloads and compares its performance to DISSECT-CF’s past behaviour. Based on the analysis, the paper concludes with recommended simulation setups that will allow the evaluation of new schedulers for peta-scale clouds in a timely fashion (e.g., within minutes). This work was partially supported by the European Union’s Horizon 2020 programme under grant agreement No 644179 (ENTICE), by the COST Program Action IC1305: Network for Sustainable Ultrascale Computing (NESUS) and by the János Bolyai Research Scholarship of the Hungarian Academy of Sciences.

Proceedings of: Second International Workshop on Sustainable Ultrascale Computing Systems (NESUS 2015). Krakow (Poland), September 10-11, 2015. This short paper introduce an algorithm for labeling connected components in n-dimensional binary images based on cellular automata, , n >= 2. Here is presented tree-dimensional binary images algorithm. The algorithm code was implemented in NetLogo programming environment. The algorithm is local and can be efficiently implemented on data-flow parallel platforms with an asymptotic complexity of O(L) on an L × L × L bynary image.

Proceedings of: Second International Workshop on Sustainable Ultrascale Computing Systems (NESUS 2015). Krakow (Poland), September 10-11, 2015. This work was supported by the European Regional Development Fund in the IT4Innovations Centre of Excellence project (CZ.1.05/1.1.00/02.0070) and the COST NESUS project with an additional CZ MEYS LD15105 support.

Proceedings of: Second International Workshop on Sustainable Ultrascale Computing Systems (NESUS 2015). Krakow (Poland), September 10-11, 2015. Current storage trends dictate placing fast storage devices in all servers and using them as a single distributed storage system. In this converged model where storage and compute resources co-exist in the same server, the role of the network is becoming more important: network overhead is becoming a main imitation to improving storage performance. In our previous work we have designed Tyche, a network protocol for converged storage that bundles multiple 10GigE links transparently and reduces protocol overheads over raw Ethernet without hardware support. However, current technology trends and server consolidation dictates building servers with large amounts of resources (CPU, memory, network, storage). Such servers need to employ Non-Uniform Memory Architectures (NUMA) to scale memory performance. NUMA introduces significant problems with the placement of data and buffers at all software levels. In this paper, we first use Tyche to examine the performance implications of NUMA servers on end-to-end network storage performance. Our results show that NUMA effects have significant negative impact and can reduce throughput by almost 2x on servers with as few as 8 cores (16 hyper-threads). Then, we propose extensions to network protocols that can mitigate this impact. We use information about the location of data, cores, and NICs to properly align data transfers and minimize the impact of NUMA servers. Our design almost entirely eliminates NUMA effects by encapsulating all protocol structures to a “channel” concept and then carefully mapping channels and their resources to NICs and NUMA nodes. We thankfully acknowledge the support of the European Commission under the 7th Framework Programs through the NanoStreams (FP7-ICT-610509) project, the HiPEAC3 (FP7-ICT-287759) Network of Excellence, and the COST programme Action IC1305, ’Network for Sustainable Ultrascale Computing (NESUS)’.

Proceedings of: Second International Workshop on Sustainable Ultrascale Computing Systems (NESUS 2015). Krakow (Poland), September 10-11, 2015. Log files are a very important set of data that can lead to useful information through proper analysis. Due to the high production rate and the number of devices and software that generate logs, the use of cloud services for log analysis is almost necessary. This paper reviews the cloud computational framework ApacheTM Hadoop R, highlights the differences and similarities between Hadoop MapReduce and Apache SparkTM and evaluates the performance of them. Log file analysis applications were developed in both frameworks and performed SQL-type queries in real Apache Web Server log files. Various measurements were taken for each application and query with different parameters in order to extract safe conclusions about the performance of the two frameworks. The authors would like to thank Okeanos the GRNET’s cloud service for the valuable resources.

Proceedings of: Second International Workshop on Sustainable Ultrascale Computing Systems (NESUS 2015). Krakow (Poland), September 10-11, 2015.

Proceedings of: Second International Workshop on Sustainable Ultrascale Computing Systems (NESUS 2015). Krakow (Poland), September 10-11, 2015. Ultrascale computing and bioinformatics are two rapidly growing fields with a big impact right now and even more so in the future. The introduction of next generation sequencing pushes current bioinformatics tools and workflows to their limits in terms of performance. This forces the tools to become increasingly performant to keep up with the growing speed at which sequencing data is created. Ultrascale computing can greatly benefit bioinformatics in the challenges it faces today, especially in terms of scalability, data management and reliability. But before this is possible, the algorithms and software used in the field of bioinformatics need to be prepared to be used in a heterogeneous distributed environment. For this paper we choose to look at sequence alignment, which has been an active topic of research to speed up next generation sequence analysis, as it is ideally suited for parallel processing. We present a multilevel stream based parallel architecture to transparently distribute sequence alignment over multiple cores of the same machine, multiple machines and cloud resources. The same concepts are used to achieve multithreaded and distributed parallelism, making the architecture simple to extend and adapt to new situations. A prototype of the architecture has been implemented using an existing commercial sequence aligner. We demonstrate the flexibility of the implementation by running it on different configurations, combining local and cloud computing resources.

Proceedings of: Second International Workshop on Sustainable Ultrascale Computing Systems (NESUS 2015). Krakow (Poland), September 10-11, 2015. Accelerators and, in particular, Graphics Processing Units (GPUs) have emerged as promising computing technologies which may be suitable for the future Exascale systems. Here, we present performance results of NekBone, a benchmark of the Nek5000 code, implemented with optimized OpenACC directives and GPUDirect communications. Nek5000 is a computational fluid dynamics code based on the spectral element method used for the simulation of incompressible flow. Results of an optimized NekBone version lead to 78 Gflops performance on a single node. In addition, a performance result of 609 Tflops has been reached on 16, 384 GPUs of the Titan supercomputer at Oak Ridge National Laboratory. This work is partially supported by EU under the COST Program Action IC1305: Network for Sustainable Ultrascale Computing (NESUS) and the Swedish e-Science Research Center (SeRC). We acknowledge PRACE for awarding us access to resource CURIE based in France at CEA as well as the computing time on the Raven system at Cray and the Titan supercomputer at Oak Ridge National Laboratory. We would also like to thank Brent Leback for the CUDA FORTRAN code used in the paper.

Proceedings of: Second International Workshop on Sustainable Ultrascale Computing Systems (NESUS 2015). Krakow (Poland), September 10-11, 2015. In this paper results of recent updates of a simple algorithm for the inversion of gravity anomalies for 3D geosections in parallel computer systems are presented. A relaxation iterative principle was used updating step by step the geosection distribution of mass density. Selection of updates was done on basis of least squares error match of the update effect with the observed anomaly. Locally weighted least squares combined with the linear trend were used to obtain good inversion results for two-body geosections.

Proceedings of: Second International Workshop on Sustainable Ultrascale Computing Systems (NESUS 2015). Krakow (Poland), September 10-11, 2015. Network traffic analysis is important for detecting intrusions and managing application traffic. Low cost, clusterbased traffic analysis solutions have been proposed for bulk processing of large blocks of traffic captures, scaling out the processing capability of a single network analysis node. Because of traffic intensity variations owing to the natural burstiness of network traffic, a network analysis cluster may have to be severely over-dimensioned to support 24/7 continuous packet block capture and processing. Bursting the analysis of some of the packet blocks to the cloud may attenuate the need for over-dimensioning the local cluster. In fact, existing solutions for network traffic analysis in the cloud are already providing the traditional benefits of cloud-based services to network traffic analysts and opening the door to cloud-based Elastic MapReduce-style traffic analysis solutions. In this paper we propose a scheduler of packet block network analysis jobs that chooses between sending the job to a local cluster versus sending it to a network analysis service on the cloud. We focus on map-intensive jobs such as string matching-based virus and malware detection. We present an architecture for an Hadoop-based network analysis solution including our scheduler, report on using this approach in a small cluster, and show scheduling performance results obtained through simulation. We achieve up to more than 50% reduction on the amount of network traffic we need to burst out using our scheduler compared to simple traffic threshold scheduler and full resource availability scheduler. Finally we discuss scaling out issues for our network analysis solution.

Proceedings of: Second International Workshop on Sustainable Ultrascale Computing Systems (NESUS 2015). Krakow (Poland), September 10-11, 2015. The long tail of science using HPC facilities is looking nowadays to instant available HPC Clouds as a viable alternative to the long waiting queues of supercomputing centers. While the name of HPC Cloud is suggesting a Cloud service, the current HPC-as-a-Service is mainly an offer of bar metal, better named cluster-on-demand. The elasticity and virtualization benefits of the Clouds are not exploited by HPC-as-a-Service. In this paper we discuss how the HPC Cloud offer can be improved from a particular point of view, of automation. After a reminder of the characteristics of the Autonomic Cloud, we project the requirements and expectations to what we name Autonomic HPC Clouds. Finally, we point towards the expected results of the latest research and development activities related to the topics that were identified. The work related to Autonomic HPC Clouds is supported by the European Commission under grant agreement H2020-6643946 (CloudLightning). The CLoudLightning project proposal was prepared by eight partner institutions, three of them as earlier partners in the COST Action IC1305 NESUS, benefiting from its inputs for the proposal. The section related to Autonomic Clouds is supported by the Romanian UEFISCDI under grant agreement PN-II-ID-PCE-2011- 3-0260 (AMICAS).

Proceedings of: Second International Workshop on Sustainable Ultrascale Computing Systems (NESUS 2015). Krakow (Poland), September 10-11, 2015. The FPGA-based accelerators and reconfigurable computer systems based on them require designing the applicationspecific processors soft-cores and are effective for certain classes of problems only, for which these soft-cores were previously developed. In Self-Configurable FPGA-based Computer Systems the challenge of designing the application-specific processors soft-cores is solved with use of the C2HDL tools, allowing them to be generated automatically. In this paper, we study the questions of the self-configurable computer systems efficiency increasing with use of the partially reconfigurable FPGAs and Chameleonc C2HDL design tool, corresponding to the goals of the project entitled "Improvement of heterogeneous systems efficiency using self-configurable FPGA-based computing" which is a part of the NESUS action. One of the features of the Chameleonc C2HDL design tool is its ability to generate a number of application-specific processors soft-cores executing the same algorithm that differ by the amount of FPGA resources required for their implementation. If the self-configurable computer systems are based on partially reconfigurable FPGAs, this feature allows them to acquire in every moment of its operation such a configuration that will provide an optimal use of its reconfigurable logic at a given level of hardware multitasking.

Proceedings of: First International Workshop on Sustainable Ultrascale Computing Systems (NESUS 2014). Porto (Portugal), August 27-28, 2014. Data in digital repositories are everyday more and more massive and distributed. Therefore analyzing them requires efficient data analysis techniques and scalable storage and computing platforms. Cloud computing infrastructures offer an effective support for addressing both the computational and data storage needs of big data mining and parallel knowledge discovery applications. In fact, complex data mining tasks involve data- and compute-intensive algorithms that require large and efficient storage facilities together with high performance processors to get results in acceptable times. In this paper we describe a Data Mining Cloud Framework (DMCF) designed for developing and executing distributed data analytics applications as workflows of services. We describe also a workflow-oriented language, called JS4Cloud, to support the design and execution of script-based data analysis workflows on DMCF. We finally present a data analysis application developed with JS4Cloud, and the scalability achieved executing it on DMCF. The work presented in this paper has been partially supported by EU under the COST programme Action IC1305, ’Network for Sustainable Ultrascale Computing (NESUS)’.

Proceedings of: First International Workshop on Sustainable Ultrascale Computing Systems (NESUS 2014). Porto (Portugal), August 27-28, 2014. One of the major challenges in ultrascale systems is the effective scheduling of complex jobs within strict timing constraints. The distributed and heterogeneous system resources constitute another critical issue that must be addressed by the employed scheduling strategy. In this paper, we investigate by simulation the performance of various policies for the scheduling of real-time directed acyclic graphs in a heterogeneous distributed environment. We apply bin packing techniques during the processor selection phase of the scheduling process, in order to utilize schedule gaps and thus enhance existing list scheduling methods. The simulation results show that the proposed policies outperform all of the other examined algorithms. The work presented in this paper has been partially supported by EU under the COST program Action IC1305, “Network for Sustainable Ultrascale Computing (NESUS)”.

Proceedings of: First International Workshop on Sustainable Ultrascale Computing Systems (NESUS 2014). Porto (Portugal), August 27-28, 2014. In this study we propose an efficient method for collaborative H.264/AVC inter-loop encoding in heterogeneous CPU+GPU systems. This method relies on specifically developed extensive library of highly optimized parallel algorithms for both CPU and GPU architectures, and all inter-loop modules. In order to minimize the overall encoding time, this method integrates adaptive load balancing for the most computationally intensive, inter-prediction modules, which is based on dynamically built functional performance models of heterogenous devices and inter-loop modules. The proposed method also introduces efficient communication-aware techniques, which maximize data reusing, and decrease the overhead of expensive data transfers in collaborative video encoding. The experimental results show that the proposed method is able of achieving real-time video encoding for very demanding video coding parameters, i.e., full HD video format, 64×64 pixels search area and the exhaustive motion estimation. This work was supported by national funds through FCT – Fundação para a Ciência e a Tecnologia, under projects PEst-OE/EEI/LA0021/2013, PTDC/EEI-ELC/3152/2012 and PTDC/EEA-ELC/117329/2010.

Proceedings of: First International Workshop on Sustainable Ultrascale Computing Systems (NESUS 2014). Porto (Portugal), August 27-28, 2014. Exascale systems are likely to have orders of magnitude less memory per core than current systems (though still large amounts of memory). As the amount of memory per core is dropping, going to thread-based models might be an unavoidable step towards the exascale milestone. AzequiaMPI is a thread-based open source full conformant implementation of MPI-1.3 for shared memory. We expose the techniques introduced in AzequiaMPI that, first, simplify the implementation and second, make the thread-based model to significantly improve the bandwidth of process-based implementations. Current version is also compliant with the MPI_THREAD_MULTIPLE thread-safety level, a feature of MPI-2.0 standard. The well known Thakur and Gropp MPI_THREAD_MULTIPLE tests show that both latency and bandwidth figures of AzequiaMPI significantly improve that of MPC-MPI, MPICH and Open MPI in an eight cores Intel Xeon E5620 Nehalem machine. The work presented in this paper has been partially supported by EU under the COST programme Action IC1305, ’Network for Sustainable Ultrascale Computing (NESUS)’.

The ever-increasing power of supercomputer systems is both driving and enabling the emergence of new problem-solving methods that require the efficient execution of many concurrent and interacting tasks. Swift/T, as a description language and runtime, offers the dynamic creation and execution of workflows, varying in granularity, on high-component-count platforms. Swift/T takes advantage of the Asynchronous Dynamic Load Balancing (ADLB) library to dynamically distribute the tasks among the nodes. These tasks may share data using a parallel file system, an approach that could degrade performance as a result of interference with other applications and poor exploitation of data locality. The objective of this work is to expose and exploit data locality in Swift/T through Hercules, a distributed in-memory store based on Memcached, and to explore tradeoffs between data locality and load balance in distributed workflow executions. In this paper we present our approach to enable locality-based optimizations in Swift/T by guiding ADLB to schedule computation jobs in the nodes containing the required data. We also analyze the interaction between locality and load balance: our initial measurements based on various raw file access patterns show promising results. Moreover, we present future work based on the promising results achieved so far. This material is based upon work supported by the U.S. Department of Energy, Office of Science, under contract DE-AC02-06CH11357. Computing resources were provided by the Argonne Leadership Computing Facility. The work presented in this paper was supported by the COST Action IC1305, “Network for Sustainable Ultrascale Computing (NESUS).” The research leading to these results has received funding from the European Union Seventh Framework Programme (FP7/2007-2013) under grant agreement number 328582.

Proceedings of: First International Workshop on Sustainable Ultrascale Computing Systems (NESUS 2014). Porto (Portugal), August 27-28, 2014. Collective operations, a key issue in the global efficiency of HPC applications, are optimized in current MPI libraries by choosing at runtime between a set of algorithms, based on platform-dependent beforehand established parameters, as the message size or the number of processes. However, with progressively more cores per node, the cost of a collective algorithm must be mainly imputed to process-to-processor mapping, because its decisive influence over the network traffic. Hierarchical design of collective algorithms pursuits to minimize the data movement through the slowest communication channels of the multi-core cluster. Nevertheless, the hierarchical implementation of some collectives becomes inefficient, and even impracticable, due to the operation definition itself. This paper proposes a new approach that departs from a frequently found regular mapping, either sequential or round-robin. While keeping the mapping, the rank assignation to the processes is temporarily changed prior to the execution of the collective algorithm. The new assignation makes the communication pattern to adapt to the communication channels hierarchy. We explore this technique for the Ring algorithm when used in the well-known MPI_Allreduce collective, and discuss the obtained performance results. Extensions to other algorithms and collective operations are proposed. The work presented in this paper has been partially supported by EU under the COST programme Action IC1305, ’Network for Sustainable Ultrascale Computing (NESUS)’, and by the computing facilities of Extremadura Research Centre for Advanced Technologies (CETACIEMAT), funded by the European Regional Development Fund (ERDF). CETA-CIEMAT belongs to CIEMAT and the Government of Spain.

Cloud-based storage is becoming a cost-effective solution for agencies, hospitals, government instances and scientific centers to deliver and share contents to/with a set of end-users. However, reliability, privacy and lack of control are the main problems that arise when contracting content delivery services with a single cloud storage provider. This paper presents the implementation of a storage system for content delivery and sharing in federated cloud storage networks. This system virtualizes the storage resources of a set of organizations as a single federated system, which is in charge of the content storage. The architecture includes a metadata management layer to keep the content delivery control in-house and a storage synchronization worker/monitor to keep the state of storage resources in the federation as well as to send contents near to the end-users. It also includes a redundancy layer based on a multi-threaded engine that enables the system to withstand failures in the federated network. We developed a prototype based on this scheme as a proof of concept. The experimental evaluation shows the benefits of building content delivery systems in federated cloud environments, in terms of performance, reliability and profitability of the storage space. The work presented in this paper has been partially supported by EU under the COST programme Action IC1305, Network for Sustainable Ultrascale Computing (NESUS).

Proceedings of: First International Workshop on Sustainable Ultrascale Computing Systems (NESUS 2014). Porto (Portugal), August 27-28, 2014. A data parallel solution approach formulated with cellular automata is proposed with a potential to become a part of future sustainable computers. It offers extreme parallelism on data-flow principles. If a problem can be formulated with a local and iterative methodology, so that data cell results always depend on neighbouring data items only, the cellular automata could be an efficient solution framework. We have demonstrated experimentally, on a graph-theoretical problem, that the performance of the proposed methodology has a potential to be for two orders of magnitude faster from known solutions.

Proceedings of: First International Workshop on Sustainable Ultrascale Computing Systems (NESUS 2014). Porto (Portugal), August 27-28, 2014. We present MOHEFT, a multi-objective list scheduling heuristic that provides the user with a set of Pareto tradeoff optimal solutions from which the one that better suits the user requirements can be manually selected. We demonstrate the potential of our method for multi-objective workflow scheduling on the commercial Amazon EC2 Cloud by comparing the quality of the MOHEFT tradeoff solutions with a state-of-the-art multi-objective approach called SPEA2* for three types of synthetic workflows with different parallelism and load balancing characteristics. We conclude with an outlook into future research towards closing the gap between the scientific simulation and real-world experimentation. The work presented in this paper has been partially supported by EU under the COST programme Action IC1305, Network for Sustainable Ultrascale Computing (NESUS).

Proceedings of: First International Workshop on Sustainable Ultrascale Computing Systems (NESUS 2014). Porto (Portugal), August 27-28, 2014. Computer systems performance is is being improved today using two major approaches: general-purpose computers computing power increase (creation of multicore processors, multiprocessor computer systems, supercomputers), and adaptation of the computer hardware to the executed algorithm (class of algorithms). Last approach often provides application of the ASIC-based and FPGA-based hardware accelerators, also called reconfigurable, and is characterized by better performance / power consumption ratio and lower cost as compared to the general-purpose computers of equivalent performance. However, such systems have typical problems. The ASIC-based accelerators: 1) are effective for certain classes of algorithms only and 2) algorithms and software require adaptation for effective application. The FPGA-based accelerators and reconfigurable computer systems (that use FPGAs as a processing unit): 1) in the process of writing require a special program to perform computing tasks balancing between the general-purpose computer and FPGAs; 2) require designing the application-specific processor soft-cores; and 3) are effective for certain classes of problems only, for which application-specific processor soft-cores were previously developed. In this paper, we consider an emerging type of high-performance computer systems called self-configurable FPGA-based computer systems, which are deprived of specified challenges. We have analyzed the background of self-configurable computer systems creation, presented current results of our research, and introduced some ongoing works. Self-configurable computer systems are being developed within the project entitled "Improvement of heterogeneous systems efficiency using self-configurable FPGA-based computing" that is the part of the NESUS Action.

Proceedings of: First International Workshop on Sustainable Ultrascale Computing Systems (NESUS 2014). Porto (Portugal), August 27-28, 2014. In this work, we study the efficiency of the OpenFOAM-based parallel solver for the heat conduction in electrical power cables. The 2D benchmark problem with three cables is used for our numerical tests. We study and compare the efficiency of conjugate gradient solver with diagonal incomplete Cholesky (DIC) preconditioner and generalized geometric algebraic multigrid solver (GAMG), which is available in Open- FOAM. The convergence and parallel scalability of the solvers are presented and analyzed. Parallel numerical tests are performed on the cluster of multicore computers. The work of authors was supported by Eureka project E!6799 POWEROPT "Mathematical modelling and optimization of electrical power cables for an improvement of their design rules". The work presented in this paper has been partially supported by EU under the COST programme Action IC1305, ’Network for Sustainable Ultrascale Computing (NESUS)’.

Proceedings of: First International Workshop on Sustainable Ultrascale Computing Systems (NESUS 2014). Porto (Portugal), August 27-28, 2014. The notion of what exactly we mean by productivity is largely depending on the active paradigms of a particular field and, on a global level, on the present prevailing social, cultural, scientific and spiritual paradigms and environment. It follows that in a long term any specific definition of productivity will have to be changed. Unfortunately, due to the historical processes, present day human-computer communication is on an extremely low level of language complexity. Consequently our present day productivity in using computers from the idea till the implementation is very low. This is primarily due to the circulus vitiosus of interdependecy of (hardware) computer architectures and popular computer programming languages based on the designs of the first Electronic Brains of the mid-last century. The natural, human Language is the prime Human tool for building a common model of the Universe, a huge fractal dynamic system, i.e. machine, whose sub-machines are smaller fractal machines consisting of a series which goes through dialects, sociolects down to idiolects. On the other hand, regarding strictly formal non-adaptable "programming" languages we see that almost all our computer linguistic efforts are oriented towards fixed expressions which are simple enough to be easily and efficiently translated into the scalar serial presently prevailing computer architecture(s). Therefore a new, fresh approach is proposed, based on the idea that the lowest possible level of a computer system shall understand a natural-like communication language, which is contextful and deals with Information, not with Data without Meta-Data. By significantly leveling up the human-computer interaction towards the ideals of a semi-natural language completely new approaches for High Productivity Computing, both on Hardware and on Software level can be thought out, and the NESUS WG1 Focus Group High Productivity Computing has been established, to historically, futuristically and realistically define and, based on that, develop, through partner collaboration projects, such a (possible) High Productivity System based on specific hardware and software. The work presented in this paper has been partially supported by EU under the COST programme Action IC1305, ’Network for Sustainable Ultrascale Computing (NESUS)’.