Computer Sciences

From AnnualReport

Jump to: navigation, search
Jesús Labarta and Eduard Ayguadé, Directors of the Computer Sciences Department

The scientific mission of the Computer Sciences Department is to influence the way computing machines are built, programmed and used. This is done through ideas, cooperation with manufacturers as technology transfer activities and “product-quality” open source developments that are usable by the scientific community. The Department includes researchers with a holistic and vertical background and vision and combines both stable and exploratory research paths, always with a co-design approach in mind, which covers from computer architecture, to resource management, performance tools, programming environments and algorithms, targeting not only supercomputer architectures but also BigData, realtime, embedded and mobile platforms. Performance, productivity, power/energy and reliability are the factors that drive our co-design approach.



Overview

The Computer Sciences Department, led by Jesús Labarta and Eduard Ayguadé, is structured in 11 research groups. Although each group has its own specialised lines of research, the teams often come together to collaborate on projects that require vertical integration, such as the Exascale EU projects Mont-Blanc/Mont-Blanc2 and DEEP/DEEP-ER, the ERC RoMoL award, the Human Brain Project (HPB) flagship and the national Severo Ochoa program. This vertical interaction is considered critical to the quality and success of the research, as feedback between the different groups enables application programmers to influence the direction of future systems architecture while better knowledge of architectures improves the design and implementation of novel programming models, execution environments and applications.


Unique Strength

The combination of broad coverage of all facets of computer systems design and programming, along with in-depth expertise in each area, are somewhat unique amongst supercomputing centres. This unique strength of the Computer Sciences Department has continued attracting leading computing companies to invest during 2014 in collaborative systems design R&D projects.


Organisational Structure

During 2014, some 190 staff and students and 50 collaborating researchers worked within the Department, organised in 11 research Groups; 4 focused on Computer Architecture (Heterogeneous Architectures, Computer Architecture for Parallel Paradigms, Operating System/Computer Architecture Interfaces and Unconventional Computer Architecture and Networks), 3 focused on improving productivity when programming and optimising parallel applications on large scale parallel systems (Programming Models, Performance Tools and Accelerators for HPC), 3 focused on APIs and resource management middleware for distributed heterogeneous architectures (Storage Systems, Grid Computing and Clusters, Autonomic Systems and e-Business Platforms) and one on novel scalable mathematical methods and algorithms for large scale systems (Extreme Computing).

Key Projects and Networks

During 2014, the Computer Sciences Department participated in the following:

- EU projects:

  • 23 FP7 projects: ASCETIC, AXLE, COMPOSE, DEEP, DEEP-ER, EESI2, EUBrazil CloudConnect, EuroServer, HOPSA-EU, Human Brain Project HBP, Lightness, Mont-Blanc, Mont-Blanc2, P-Socrates, ParaDIME, parMERASA, PROXIMA, ReNewIT, RETHINK Big, SECURED, TERAFLUX, TransPLANT and ARTEMIS VeTeSS;
  • RoMoL (Riding on Moore's Law) ERC grant (Mateo Valero);
  • Hi-EST (Holistic Integration of Emerging Supercomputing Technologies) ERC grant (David Carrera);
  • The Second Implementation Phase in the PRACE EU FP7 project: FTI and energy-efficient prototype;
  • The HiPEAC-3 network of excellence;
  • The European Technology Platform for HPC (ETP4HPC);
  • The SCALUS Marie Curie Initial Training network.

- Collaborations with IT companies:

  • Intel Corporation with a multi-year agreement Intel-BSC Exascale Lab, on topics related to analysis of applications and performance prediction tools and programming models;
  • NVIDIA through the CUDA Center Of Excellence (CCoE), in association with the Universitat Politècnica de Catalunya (UPC-BarcelonaTECH);
  • Microsoft Research through the BSC-Microsoft Research Centre agreement, on topics related to low-power vector architectures, architectural support for programming languages and analysis of Hadoop for MapReduce workloads;
  • IBM Research through the following Joint Study Agreements: High-performance in-memory databases, Software-defined environments for HPC workloads, Adaptive resource management for Power, OmpSs @ P8/GPU, OmpSs programming model for asynchronous applications, Resilience compiler support, Performance API for OpenMP and Applicable research to interconnection networks;
  • Samsung to evaluate memory behavior of HPC production applications and frequency and locality of memory errors;
  • Xilinx Ireland to research on OmpSs support for FPGA accelerators;
  • Qualcomm to port BSC's instrumentation packages to their DragonBoard development kit;
  • The European Space Agency (ESA), with the “Multicore OS benchmark” and "Architectural solutions for the timing predictability of next-generation multi-core processors" project;

- National project:

  • The BSC-CNS Severo Ochoa program, where the Department is developing a novel platform to support interactive simulation and computational workflows with Big Data requirements, to be applied to the three challenging applications in the project from Life and Earth Sciences and CASE Departments.


Scientific Output

For additional information, please see the Detailed Report of Research Activities 2014.


Impacting the Future of Computing
In collaboration with market leaders such as IBM, Microsoft, Intel and NVIDIA, as well as other international computing centres and standardisation efforts, the researchers of the Computer Sciences Department are involved in a range of projects covering the full spectrum of next generation computer design, from novel processor and multicore architectures, energy-efficient systems based on mobile components, architectural support to the software stack (e.g. runtime systems and OS), programming and execution models (e.g. OpenMP and StarSs), as well as support for the efficient programming and management of BigData and Cloud architectures. Some of the results of their work are considered a reference worldwide.



The research results of the Department were published in the proceedings of high quality conferences in the area (A* and A in CORE2013 conference ranking, www.core.edu.au/coreportal), including International Symposium on Computer Architecture (ISCA), International Conference on Parallel Architectures and Compilation Techniques (PACT), Annual Design Automation Conference (DAC), International Conference on Supercomputing (ICS), International Parallel & Distributed Processing Symposium (IPDPS), Design, Automation, and Test in Europe (DATE) conference, International Conference on Computing Frontiers (CF), International Conference on Hardware/Software Codesign and System Synthesis (CODES+ISSS), International Conference on Application Specific Systems, Architectures, and Processors (ASAP), Symposium on Principles and Practice of Parallel Programming (PPoPP), International Conference on Parallel Processing (ICPP), International Symposium on Cluster Computing and the Grid (CCGrid) and International Conference on Cluster Computing (CLUSTER).

In addition, other more consolidated research results have been published in prestigious journals in the area, including IEEE Transactions on Computers, IEEE Micro, IEEE Transactions on Parallel and Distributed Systems, IEEE Transactions on Very Large Scale Integration Systems, ACM Transactions on Architecture and Code Optimization, ACM Transactions on Design Automation of Electronic Systems, Future Generation Computing Systems, Concurrency and Computation: Practice and Experience, Parallel Computing, Journal of Parallel and Distributed Computing, International Journal of Parallel Programming and Information Systems Frontiers.

Nine PhD thesis were defended during 2014 in the Computer Architecture Department at the Universitat Politècnica de Catalunya (UPC-BarcelonaTECH).


Research Groups

Heterogeneous Architectures

This Group aims to design and evaluate the next generation of HPC systems capable of achieving the energy efficiency required by future Exaflop supercomputers. The Group believes that components coming from the embedded and mobile device markets will replace current HPC components coming from the PC and server markets, due to their larger unit count, lower unit price, rapid evolution, and higher energy efficiency. To that end, during 2014, the Group performed Tracked the evolution of ARM-based chips from either the mobile market, or the server market and successfully installed the first chassis of the Mont-Blanc prototype consisting of 135 compute nodes based on Samsung Exynos 5 dual SoC interconnected through gigabit ethernet with a total power consumption below 3 kWatts. The Group also participated in the FP7 EuroServer project, evaluating the use of energy-efficient technologies from the mobile market for BigData processing, and continued the collaboration with Samsung Co., Ltd. in the field of memory systems for high-performance computing (HPC). Finally, the Group developed performance and power modeling tools for next-generation HPC systems with a special focus at the node level with the objective of predicting the impact of architectural enhancements in the compute units. The Group was lead by Alex Ramirez during the first half of the year; since then the Group is managed by the Department Directors.



Computer Architecture for Parallel Paradigms

Led by Adrián Cristal and Osman Unsal, the Group focuses its research on the architectural support for novel programming models and execution environments for future multicore architectures. The Group focuses its research on lowering the programmability wall raised by multicore architectures; research areas include low-power vector processors, transactional memory, reliability and hardware for big data. During 2014, the group continued its work in five FP7 projects (ParaDIME, AXLE, RETHINK big, ICT-Energy and Mont Blanc). The group finalized their work on reliability as part of the Intel-BSC Laboratory and the G8 Enabling Climate Simulation at Extreme Scale project. In April 2014, the group also finalized their work in the BSC-Microsoft Research Centre, and passed the baton to the BSC Autonomic Systems and eBusiness Platforms research group to continue the center activities.


Operating System / Computer Architecture Interface

The CAOS Group has focused its research on embedded real-time and high-performance systems. During 2014, in the area of real-time systems the group successfully completed the EU FP7 parMERASA project, started three projects with the European Space Agency and continued its participation in the EU FP7 PROXIMA, P-SOCRATES and ARTEMIS VeTeSS projects. In the area of high-performance systems, the group continued its work on energy and CPU accounting and on adaptive prefetching techniques as part of a Joint Study Agreement with IBM. Researchers of the Group constitute the core of the research activities in the RoMoL ERC grant.


Unconventional Computer Architecture and Networks

Led by Mario Nemirovsky, UCAN group is conducting research on Internet of Things (IoT), Fog computing, Big Data, simulation techniques, and processor architectures. In this direction, the group has been positioning as a referent in platforms for IoT. Areas such as security, virtualization, and infrastructure are under study, proposing new solutions to break the silos problem while maintaining real-time constraints. The group is also working on new processor architectures and algorithms that can take advantage of Wireless Network-on-Chip (NoC), by using antennas at core-level to enable a natural broadcast network to communicate the cores. Additionally, UCAN group is consolidating new simulation techniques based on queue models, allowing faster simulations with an excellent accuracy.


Programming Models

Led by Xavier Martorell, the Group explores new programming models and their efficient implementation for current and future architectures, from manycore SMPs with support for accelerators (GPUs, MIC, FPGAs), to clusters of SMPs with accelerators, to exascale systems. The research is supported by two powerful tools: the Mercurium compiler, used to prototype new programming model approaches, and Nanos++, the runtime library supporting the variety of hardware resources under consideration. During 2014, the Group improved the support for GPUs with CUDA and OpenCL, with the use of an asynchronous thread to deal with the GPU operations. The Group also worked on the support for FPGAs and developed a preliminary implementation on OmpSs using the hStream model for the MIC processor. The Group also continued the development of DLB (Dynamic Load Balancing) techniques and providing a better interoperability solutions between MPI and OmpSs, and OpenMP. The Group developed DSL (Domain Specific Languages) extensions for solving partial diferential equations as part of the collaboration with Repsol. The Group continued the work on power modelling for multicore architectures and code transformations for processors with local memories. During 2014, the Group also continued the participation in the EU FP7 projects Mont-Blanc (I and II), DEEP, DEEP-ER, the Human Brain Project (HBP) and TERAFLUX (finalized), as well as in the HiPEAC-3 network of excellence and the research collaborations with Intel, IBM and Xilinx Research Labs in Dublin. Finally, the Group successfully applied for a H2020 project, AXIOM, on research on new software/hardware architectures for Cyber-Physical Systems.


Performance Tools

This Group, led by Judit Giménez, works on the design of tools to measure, analyze and predict the behavior of parallel applications on parallel systems. The main goal of the Group aims at providing technologies to understand the issues that determine the actual performance of a parallel application or contribute to the application bottlenecks. This knowledge is extremely important in both novel homogeneous and heterogeneous multi-core architectures as well as in highly scalable cluster systems. During 2014, the activities developed within the Group evolved the research lines targeting performance analytics and models as well as improving the tools infrastructures and their integration. The Group also participated in the EU exascale projects Mont-Blanc/Mont-Blanc2 and DEEP/DEEP-ER. With respect to projects with private companies, the group is highly involved in the joint Intel-BSC Exascale Laboratory and during this year we also had a project with IBM. We have established an international collaboration with RIKEN (K-Computer) as well as contributed to the JLESC activities.


Grid Computing & Clusters

Led by Rosa M. Badia, this Group is researching new programming and execution models and resource management techniques for distributed computing. The Group explores solutions in order to simplify application development, to enable dynamic exploitation of parallelism at runtime, and to perform combined scheduling decisions at different levels. In these directions, the efforts of the Group during 2014 focused on further development of the COMPSs programming model, specially in the Python binding, the completion of the graphical IDE to support application developers, support to multiple task versions, new implementation of the support to tasks' constraints, and other updates.

The Group has also continued with the integration of COMPSs with novel storage technologies, like the dataClay self-contained objects layer or the Hecuba layer for Cassandra databases, both in development as part of the Severo Ochoa programme). The Group is also researching new technologies for low-cost energy efficient microservers. Finally, the Group has continued its participation in several EU FP7 projects (transPLANT, HBP, ASCETIC, EuroServer and EUBrazil CloudConnect) as well as in the Severo Ochoa Centre of Excellence, all of them related with extensions of COMPSs and porting of applications.


Autonomic Systems and eBusiness Platforms

Led by Jordi Torres, this group performs high-level research in eBusiness applications and platforms executing on high-productivity multiprocessor architectures as well as distributed environments. The research group explores the future of computing by performing high-level research in today’s eBusiness applications that have to deal with critical IT challenges in areas such as Cognitive Computing, Big Data, Cloud Computing, Business Analytics, High Performance Computing and Sustainable Computing. The goal of this team is to contribute in the continuous development of supercomputing systems enabling the convergence of advanced analytic algorithms and big data technologies driving new insights based on the massive amounts of available data. The group is conducting research on autonomic and intelligent resource management policies based on Self-Management strategies as the way to improve the computer middleware layers. The Group is also researching new architectural proposals for the memory/storage hierarchy including processing-in-memory techniques and novel key-value storage models to support Big Data workloads. The research is supported by different projects as the EU FP7 COMPOSE, RenewIT, ASCETiC, EuroServer and LIGHTNESS projects; the Severo Ochoa program; a collaboration project with Microsoft; a collaboration project with CA and two collaborations with IBM Research.



Storage Systems

Led by Toni Cortés, this Group explores appropriate solutions to the scalability of parallel storage systems in large installations (in which very large volumes of data need to be generated and accessed), new file-system approaches to increase their performance, and new approaches to store and manage Big Data. In 2014, the Group focussed on increasing the performance and scalability of parallel file systems by applying de ideas of guided IO (hints by the programmer) and extending the direct look up file system designed in 2012 to support new key-value devices. In addition, the group has continued the development a new storage abstraction based on self-contained objects. During 2014 the research activities of the Group have been supported by the Severo Ochoa programme, the Mont-Blanc, PRACE and HBP EU projects.


Extreme Computing

Led by Vassil Alexandrov, this Group focuses on development of novel scalable mathematical methods and algorithms for large scale systems and applying these to solving problems with uncertainty on such systems. The Group's main expertise is in the area of computational science, novel scalable mathematical methods and algorithms for large scale systems and exascale computing paradigm. In particular, scalable Monte Carlo and hybrid algorithms are developed for linear algebra, optimisation, computational finance, environmental models, computational biology, etc. In addition, the research focuses on scalable, fault-tolerant and resilient algorithms for extreme scale (petascale and exascale) computing. During 2014 the Group has been recognised as emerging research group in Extreme Scale Computing by the Catalan Government.


Personal tools
Home