Swedish web site
SaS Seminars 2014
Software and Systems Research Seminar Series
The SaS Seminars are a permanent series of open seminars of the Division of Software and Systems (SaS) at the Department of Computer and Information Science (IDA), Linköping University. The objective of the seminars is to present outstanding research and ideas/problems relevant for SaS present and future activities. In particular, seminars cover the SaS research areas software engineering, programming environments, system software, embedded SW/HW systems, computer systems engineering, realtime systems, parallel and distributed computing, and theoretical computer science. - Two kinds of seminars are planned:
talks by invited speakers not affiliated with SaS,
internal seminars presenting lab research to whole SaS.
The speakers are expected to give a broad perspective of the presented research, adressing the audience with a general computer science background but possibly with no specific knowledge in the domain of the presented research. The normal length of a presentation is 60 minutes, including discussion.
The SaS seminars are coordinated by Christoph Kessler.
SaS Seminars (2014)
Recent contributions to energy efficient computing systems by the CARD-group at NTNU
Prof. Lasse Natvig, NTNU Trondheim, Norway
Tuesday, 16 Dec. 2014, 10:15, room Alan Turing
Abstract:
The Computer Architecture and Design Group (CARD) at NTNU in Trondheim has in the last 7 years put an increasing focus on energy efficiency in its research within hardware, software, and in computer systems in general. This research spans from basic research in unconventional computing to industrially relevant experiments. The presentation will describe the industrially relevant activity within energy efficient computing systems, ranging from the micro architectural level via compilers and system software to experiments on HPC platforms. A brief overview and our vertical approach will be introduced before we go into several examples from recent research projects and publications.
Speaker's bio:
Lasse Natvig is professor of computer architecture since 1997.
He worked 4 years for Tandberg Data in Oslo during the 80'ies,
and has a PhD degree in Parallel algorithms from 1991.
Natvig has since then been at the Computer Architecture and Design group
of Department of Computer and Information Science at NTNU.
He has supervised close to 100 students in projects and thesis work
within computing systems, most of high industrial relevance.
Several of these students have started companies during their study
or just after their study.
Natvig's current research interests are computer architecture,
parallel programming and energy efficient computing systems.
Natvig is full member of HiPEAC NoE.
Accurate Energy Modelling for Many-Core Static Schedules
Prof. Jörg Keller, FernUniv. in Hagen, Germany
Tuesday, 18 Nov. 2014, 13:15, room Alan Turing
Abstract:
Static schedules can be a preferable alternative for
applications with timing requirements and predictable behaviour
since the resources can be more precisely allocated for the given
workload. Unused resources are handled by power management
systems to either scale down or shut off parts of the chip to save
energy. In order to efficiently implement power management,
especially in many-core systems, an accurate model is important
in order to drive the appropriate power management decisions
at the right time. Practical issues such as latency for driving
the power saving techniques should hence be considered when
deriving the system model, especially for fine timing granularity.
In this paper we present an accurate energy model for
many-core systems taking into account practical aspects such
as switching latency of modern power saving techniques. The
model is used to calculate an optimal static schedule for manycore
task execution on systems with dynamic frequency levels
and sleep state mechanisms implemented. By using the energy
model, the coordination between power management and task
execution creates an energy optimal system without performance
degradation. We derive all model parameters for an embedded
processor, and validate the model with synthetic benchmarks
on this processor, demonstrating its accuracy and forecasting
capability.
Speaker's profile:
Jörg Keller is professor at the Faculty of Mathematics and
Computer Science at
FernUniversität in Hagen, Germany, where
he is chair of the group
Parallelism and VLSI.
His research interests include Parallel Computing, Security and Cryptography,
Fault Tolerant Computing, and Online Learning.
He holds a PhD degree and a habilitation degree from Universität des
Saarlandes, Saarbrücken, Germany.
Addressing the Exascale Challenge from an Application Viewpoint
Prof. Erwin Laure, KTH/PDC, Stockholm
Tuesday, 16 Sep. 2014, 16:15, room Alan Turing
Abstract:
For the past thirty years, the need for ever greater supercomputer performance has driven the development of many computing technologies which have subsequently been exploited in the mass market. Delivering an exaflop (or 10^18 calculations per second) by the end of this decade is the challenge that the supercomputing community worldwide has set itself. A range of technological innovations are needed to build viable (and affordable!) exascale systems but even if they exist, there is still the question of how to program them and what applications can use them efficiently.
In this talk we report on two projects that tackle these issues:
The Collaborative Research into Exascale Systemware, Tools and Applications
project (CRESTA) focuses on the use of six applications with exascale potential
and uses them as co-design vehicles to develop: the development environment,
algorithms and libraries, user tools, and the underpinning and cross-cutting
technologies required to support the execution of applications at the exascale.
The applications represented in CRESTA have been chosen as a representative
sample from across the supercomputing domain including: biomolecular systems,
fusion energy, the virtual physiological human, numerical weather prediction
and engineering.
The Exascale Programming Models (EPiGRAM) project is addressing the exascale
challenge by improving one of the most widely used programming models,
message passing, considering also the impact of PGAS approaches.
We will particularly focus on initial findings on scalability limits,
and propose directions to overcome them.
Vita:
Erwin Laure is Professor in Computer Science,
Head of the Department for High Performance Computing (HPCViz),
and Director of PDC - Center for High Performance Computing
at the Royal Institute of Technology (KTH) in Stockholm, Sweden.
He is the Coordinator of the EC-funded project EPiGRAM
investigating programming models for exascale computing with a
focus on message passing and PGAS, and actively involved in major European
HPC infrastructure projects (EGI, PRACE, EUDAT) for many years.
He holds a PhD in Business Administration and Computer Science
from the University of Vienna, Austria.
His research interests include programming environments, languages, compilers
and runtime systems for parallel and distributed computing,
particularly exascale computing as well as grid and cloud computing.
Optimizing OpenCL for ARM Mali GPUs
Johan Grönqvist, ARM, Lund
Monday, 1 Sep. 2014, 10:15, room Grace Hopper
Abstract:
The GPU architecture of ARM Mali GPUs will be presented,
as well as the hardware counters that can be used to understand
the performance characteristics of a kernel executing on the GPU.
Following that, a workflow will be described for detecting
bottlenecks and finding optimization opportunities.
If there is time left, an example will be discussed in which
a matrix multiplication kernel is optimized (and memory access
patterns play an important role).
Speaker's bio:
Johan Grönqvist is a Senior Graphics Software Engineer at ARM, Lund.
His recent work involved analysing and improving compute performance
on ARM GPUs. Tasks have involved the design of a 2D graphics driver,
benchmarking and performance analysis, as well as implementation of
the math library. For the past 18 months, he has been part of the
compiler team, where he is now again focusing on performance analysis of GPUs.
Type-Based Structural Analysis for Modular Systems of Equations
Dr. Henrik Nilsson, Univ. of Nottingham, UK
Thursday, 19 June 2014, 09:00 (sharp), room John von Neumann
Abstract:
Descriptions of phenomena and entities in terms of systems of equations play a
prominent role in many branches of science and engineering. Computers are then
routinely used to facilitate the study of the object of interest. A typical
example is modelling and simulation of physical systems. The systems of
equations can be very large, making a modular formulation a practical
necessity to allow reuse, facilitate maintenance, and so on. It then becomes
desirable to know if an isolated equation system fragment is well-formed, or
if it has some inherent flaw that makes it impossible to use as a part of
meaningful system of equations. In this talk, I will consider
structural properties of systems of equations that can help answering
this question, and how to capture such properties in the type of
of an equation system fragment so as to support a compositional analysis.
Short bio:
Dr. Henrik Nilsson is a lecturer at the School of Computer Science,
The University of Nottingham, where he works in the Functional Programming Lab.
His research interests include functional programming,
functional programming environments, and various aspects of design,
semantics, and implementation of programming languages in general,
such as advanced type systems and their applications. He is also
interested in modelling and simulation languages, and how such
languages can be improved by using techniques developed in the
declarative language community. He got a PhD in 1998
from Linköping university,
where he worked at the Programming Environments Laboratory
on debugging tools for lazy functional languages.
Ebba: An Embedded DSL for Bayesian Inference
Dr. Henrik Nilsson, Univ. of Nottingham, UK
Tuesday, 17 June 2014, 13:15, room Alan Turing
Abstract:
A probabilistic model describes a stochastic process or system
in terms of a probability distribution for the outcomes given
the parameters of the model.
The typical situation is that the outcomes can be observed directly,
whereas the parameters cannot, and thus have to be estimated based
on some particular observed outcomes.
Bayesian inference is a principled way to estimate parameters of
probabilistic models given observed data.
Roughly, exploiting Bayes theorem, the probabilistic model is "inverted",
yielding a probability distribution for the parameters given the observations.
A number of probabilistic programming languages exists that allow
probabilistic models to be implemented in a way that models can
be "inverted" automatically, allowing Bayesian inference to be carried out,
typically through a probabilistic algorithm.
However, these languages are stand-alone implementations.
In this talk, I will investigate the possibility of an embedded implementation
of such a probabilistic language supporting Bayesian inference.
The starting point is Baysig, a Haskell-like language developed by
OpenBrain Ltd, where a probabilistic computation is represented by a monad.
A model conditioned on parameters is thus a function returning a monad,
and, through the Kleisli construction, an arrow.
This suggests that something like arrows might be an appropriate abstraction
for building an "invertible" model, as arrows explicitly relate inputs to outputs.
This was the starting point for Ebba, short for Embedded Baysig.
While still very preliminary work, the initial experience with Ebba is promising.
Short bio:
See above
Parallel Computer Research at DTU - Some Highlights
Dr. Sven Karlsson, DTU, Denmark
Monday, 16 June 2014, 13:15, room Alan Turing
Abstract:
Computer architecture activities at DTU have grown over the last five years
and now include work on all levels of the software stack as well as
micro-architecture and hardware realization.
This talk will cover both a brief overview of all activities and some
highlights from our work on parallel middle-ware, performance debugging
tools, hardware realization and design.
Short bio:
Dr. Sven Karlsson is associate professor at
the Technical University of Denmark (DTU),
Department of Applied Mathematics and Computer Science,
Section for Computer Science and Engineering,
in Lyngby, Denmark.
His research interests include
programming models, parallel computers, compilers, operating systems,
computer networks and computer architecture.
He holds a PhD in teleinformatics from KTH, Stockholm.
Structured parallel programming models: the FastFlow experience
Dr. Marco Danelutto, Dept. of Computer Science, Univ. of Pisa, Italy
Thursday, 8 May 2014, 10:15, room Alan Turing
Abstract:
FastFlow is the structured parallel programming framework
developed and maintained at the Univ. of Pisa and Torino.
Originally aimed at providing efficient support for stream
parallel computations only on shared memory multi core
architectures, it has been extended (a) to cover data parallel
computations and (b) to target heterogeneous machines
(multi core + GPU), many core architectures (Tilera and Xeon PHI)
and networked heterogeneous machines (COW/NOW).
FastFlow is built on top of POSIX Pthreads and C++, and in
the most recent versions exploits different C++11 features
to provide enhanced programmability. Two key features distinguish
FastFlow from similar programming frameworks, namely the ultra
efficient, lock and fence free communication mechanisms supporting
the implementation of very efficient fine grain parallel computations,
and the possibility to implement "software accelerators"
using spare cores (or many core attached boards) in the
architecture to accelerate existing sequential code.
We describe the parallel programming framework salient features
and we discuss different points giving different perspectives
to structured parallel programming: implementation of alternative
programming models (in particular data flow) with the existing
skeletons, support for management of non functional features
typical of parallel programming (performance, resilience,
power management, etc.), and eventually RISC-pbb, a set of
parallel building blocks used in FastFlow to implement all the
patterns provided and general enough to support the implementation
of different parallel patterns and programming models.
Short Bio:
Marco Danelutto is an associate professor at the University of Pisa.
He has been and is active in the parallel programming model and
tool area since early '90s. His research has concentrated in particular
on structured parallel programming models, such as those based on
algorithmic skeletons and parallel design patterns.
He has been one of the main designers of P3L, the skeleton language
developed in Pisa in the '90s, he contributed to the development
of the Behavioural skeleton model within CoreGrid (EU FP6 NoE)
and GridCOMP (EU FP6 STREP) in the early '00s and more recently
he has been involved in the design and development of the FastFlow
parallel programming framework (used within the FP7 STREP projects
ParaPhrase and REPARA). He first introduced the techniques exploiting
macro data flow technology in algorithmic skeleton framework
implementation (late '90s) and the concept of autonomic manager
taking care of non-functional concerns in structured parallel
program execution (early '00s). He is author and co-author of
about 140 papers in refereed international conferences and journals.
Collaborative Robotics
Dr. Alexander Kleiner, IDA, Linköping University
Thursday, 27 february 2014, 10:15, room Alan Turing
Abstract:
Increasingly cheaper computer technology, as well as sensor and
actuator systems in robotics today are paving the way for large teams
of collaborating robots. The coordination of large robot teams leads
to almost intractable combinatorial problems as they were never relevant
in practice before. Therefore, there exists an increasing demand for
time-efficient approaches that are capable of solving heavy
combinatorial problems as they appear in robotics and multi-agent
systems today. Such problems arise, for example, in the application
domains of manufacturing and intra-logistics where numerous mobile
robots need to actively collaborate for managing transportation tasks.
Also in search and rescue (SAR) robot coordination becomes
computationally challenging with larger robot teams searching for either
stationary or mobile targets, for example, when coordinating a team of
unmanned aerial vehicles (UAVs) searching for lost hikers in the Alps.
In this talk I will provide an overview on cognitive methods that I
developed during the last years for facilitating successful
collaboration in robot teams. I will provide examples from two target
domains which are collaborative robots handling transportation tasks in
intra-logistics, and teams of UAVs searching for survivors in Search and
Rescue.
Short Bio:
Alexander Kleiner is docent and university lecturer at the computer
science department (IDA) at the Linköping University. He obtained his
docent degree in December 2013 from the Linköping University and his
Ph.D. degree (Dr. rer. nat) from the University of Freiburg in February
2008. He worked as an invited guest researcher at the Carnegie Mellon
University, Pittsburgh, USA in 2010 and at the La Sapienza University,
Rome, Italy in 2011. Since 2006, he is member of the executive committee
of RoboCup (Rescue Simulation League) and since 2008 member of the IEEE
Technical Committee on Safety Security and Rescue Robotics. He served as
General Chair of the IEEE International Symposium on Safety, Security,
and Rescue Robotics 2013 and program chair in 2012.
His research area focuses on collaborative robotics including autonomous
robot exploration, guaranteed search, simultaneous localization and
mapping (SLAM), distributed task allocation, and multi-robot motion
planning. He published more than 70 papers and received several
scientific awards. He successfully participated in several international
robot competitions where his teams won almost constantly the first
prize. Besides his research and teaching activities at the university,
he works as a consultant for the industry where he works on projects
implementing fleets of autonomous mobile robots for solving
transportation tasks in intra-logistics and production.
Previous SaS Seminars
Page responsible: Christoph Kessler
Last updated: 2016-03-01
