Technical Reports

A List by Author: Lubo¹ Brim

e-mail:

brim(a)fi.muni.cz

home page:

https://www.fi.muni.cz/usr/brim/

Computing Strongly Connected Components in Parallel on CUDA (full version)

by Jiøí Barnat, Petr Bauch, Lubo¹ Brim, Milan Èe¹ka, July 2010, 24 pages.

FIMU-RS-2010-10. Available as Postscript, PDF.

Abstract:

The problem of decomposition of a directed graph into its strongly connected components is a fundamental graph problem inherently present in many scientific and commercial applications. In this paper we show how existing parallel algorithms can be reformulated in order to be accelerated by NVIDIA CUDA technology. In particular, we design a new CUDA-aware procedure for pivot selection and we redesign the parallel algorithms in order to allow for CUDA accelerated computation. We also experimentally demonstrate that with a single GTX 280 GPU card we can easily outperform optimal serial CPU implementation, which is particularly interesting result as unlike the serial CPU case, the asymptotic complexity of the parallel algorithms is not optimal.

Using Strategy Improvement to Stay Alive

by Lubo¹ Brim, Jakub Chaloupka, March 2010, 48 pages.

FIMU-RS-2010-03. Available as Postscript, PDF.

Abstract:

We design a novel algorithm for solving Mean-Payoff Games (MPGs). Besides solving an MPG in the usual sense, our algorithm computes more information about the game, information that is important with respect to applications. The weights of the edges of an MPG can be thought of as a gained/consumed energy -- depending on the sign. For each vertex, our algorithm computes the minimum amount of initial energy that is sufficient for player Max to ensure that in a play starting from the vertex, the energy level never goes below zero. Our algorithm is not the first algorithm that computes the minimum sufficient initial energies, but according to our experimental study it is the fastest algorithm that computes them. The reason is that it utilizes the strategy improvement technique which is very efficient in practice.

Faster Algorithm for Mean-Payoff Games

by Jakub Chaloupka, Lubo¹ Brim, October 2009, 11 pages.

FIMU-RS-2009-08. Available as Postscript, PDF.

Abstract:

We study some existing techniques for solving mean-payoff games (MPGs), improve them, and design a randomized algorithm for solving MPGs with currently the best expected complexity.

CUDA accelerated LTL Model Checking

by Jiøí Barnat, Lubo¹ Brim, Milan Èe¹ka, Tomá¹ Lamr, June 2009, 18 pages.

FIMU-RS-2009-05. Available as Postscript, PDF.

Abstract:

Recent technological developments made available various many-core hardware platforms. For example, a SIMD-like hardware architecture became easily accessible for many users who have their computers equipped with modern NVIDIA GPU cards with CUDA technology. In this paper we redesign the maximal accepting predecessors algorithm [7] for LTL model checking in terms of matrix-vector product in order to accelerate LTL model checking on many-core GPU platforms. Our experiments demonstrate that using the NVIDIA CUDA technology results in a significant computation speedup.

Partial Order Reduction for State/Event LTL

by Nikola Bene¹, Lubo¹ Brim, Ivana Èerná, Jiøí Sochor, Pavlína Vaøeková, Barbora Zimmerová, July 2008, 21 pages.

FIMU-RS-2008-07. Available as Postscript, PDF.

Abstract:

Software systems assembled from a large number of autonomous components become an interesting target for formal verification due to the issue of correct interplay in component interaction. State/event LTL incorporates both states and events to express important properties of component-based software systems.

The main contribution of the paper is a partial order reduction technique for verification of state/event LTL properties. The core of the partial order reduction is a novel notion of stuttering equivalence which we call state/event stuttering equivalence. The positive attribute of the equivalence is that it can be resolved with existing methods for partial order reduction. State/event LTL properties are, in general, not preserved under state/event stuttering equivalence. To this end we define a new logic, called weak state/event LTL, which is invariant under the new equivalence.

LTL model checking with I/O-Efficient Accepting Cycle Detection

by Jiøí Barnat, Lubo¹ Brim, Pavel ©imeèek, January 2007, 20 pages.

FIMU-RS-2007-01. Available as Postscript, PDF.

Abstract:

We show how to adopt existing non-DFS-based algorithm OWCTY for accepting cycle detection to the I/O efficient setting and compare the I/O efficiency and practical performance of the adopted algorithm to the existing I/O efficient LTL model checking approach of Edelkamp et al. We show that while the new algorithm exhibits similar I/O complexity with respect to the size of the graph, it avoids the quadratic increase in the size of the graph of the approach of Edelkamp et al. Therefore, the absolute numbers of I/O operations are significantly smaller and the algorithm exhibits better practical performance.

Distributed Qualitative LTL Model Checking of Markov Decision Processes

by Jiøí Barnat, Lubo¹ Brim, Ivana Èerná, Milan Èe¹ka, Jana Tùmová, September 2006, 19 pages.

FIMU-RS-2006-04. Available as Postscript, PDF.

Abstract:

Probabilistic processes are used to model concurrent programs that exhibit uncertainty. The state explosion problem for probabilistic systems is more critical than in the non-probabilistic case. In the paper we propose a cluster-based algorithm for qualitative LTL model checking of finite state Markov decision processes. We use the automata approach which reduces the model checking problem to the question of existence of an accepting end component. The algorithm uses repeated reachability which systematically eliminates states that cannot belong to any accepting end component. A distinguished feature of the distributed algorithm is that its complexity meets the complexity of the best known sequential algorithm.

Under-Approximation Generation using Partial Order Reduction

by Lubo¹ Brim, Ivana Èerná, Pavel Moravec, Jiøí ©im¹a, A full version of the paper submitted to conference CAV05 February 2005, 21 pages.

FIMU-RS-2005-04. Available as Postscript, PDF.

Abstract:

We propose a new on-the-fly approach which combines partial order reduction with the under-approximation technique for falsification and verification of LTL-X properties. It uses sensitivity relation and modified ample conditions to generate a reduced state space that is not fully stutter equivalent to the original one and it checks the desired property using representatives. Widening of under-approximations is fully automatic and does not rely on any supporting mechanisms like theorem-provers or SAT solvers.

Accepting Predecessors are Better than Back Edges in Distributed LTL Model Checking

by Lubo¹ Brim, Ivana Èerná, Pavel Moravec, Jiøí ©im¹a, A full version of the paper accepted to the conference FMCAD 2004. November 2004, 22 pages.

FIMU-RS-2004-09. Available as Postscript, PDF.

Abstract:

We present a new distributed-memory algorithm for enumerative LTL model-checking that is designed to be run on a cluster of workstations communicating via MPI. The detection of accepting cycles is based on computing maximal accepting predecessors and the subsequent decomposition of the graph into independent predecessor subgraphs induced by maximal accepting predecessors. Several optimizations of the basic algorithm are presented and the influence of the ordering on the algorithm performance is discussed. Experimental implementation of the algorithm shows promising results.

Distributed Memory LTL Model Checking Based on Breadth First Search

by Jiøí Barnat, Lubo¹ Brim, Jakub Chaloupka, September 2004, 57 pages.

FIMU-RS-2004-07. Available as Postscript, PDF.

Abstract:

We propose a parallel distributed memory on-the-fly algorithm for enumerative LTL model checking. The algorithm is designed for network of workstations communicating via MPI. The detection of cycles (faulty runs) effectively employs the so called back-level edges. In particular, a parallel level synchronized breadth first search of the graph is performed to discover all back-level edges and for each level the back-level edges are checked in parallel by a nested search procedure to confirm or refute the presence of a cycle. Several optimizations of the basic algorithm are presented and advantages and drawbacks of their application to distributed LTL model checking are discussed. Experimental evaluation of the algorithm is presented.

Parallel Algorithms for Detection of Negative Cycles

by Lubo¹ Brim, Ivana Èerná, Luká¹ Hejtmánek, This is a full version of the paper presented at PARCO 2003. July 2003, 14 pages.

FIMU-RS-2003-04. Available as Postscript, PDF.

Abstract:

Several new parallel algorithms for the single source shortest paths and for the negative cycle detection problems on directed graphs with real edge weights and given by adjacency list are developed, analysed, and experimentally compared. The algorithms are to be performed on clusters of workstations that communicate via a message passing mechanism.

Using Assumptions to Distribute CTL Model Checking

by Lubo¹ Brim, Jitka Crhová, Karen Yorav, This is a full version of the paper presented at PDMC`02. October 2002, 22 pages.

FIMU-RS-2002-08. Available as Postscript, PDF.

Abstract:

In this work we discuss the problem of performing distributed CTL model checking by splitting the given state space into several "partial state spaces". The partial state space is modelled as a Kripke structure with border states. Each computer involved in the distributed computation owns a partial state space and performs a model checking algorithm on this incomplete structure. To be able to proceed, the border states are augmented by assumptions about the truth of formulas and the computers exchange assumptions about relevant states as they compute more precise information. In the paper we give the basic definitions and present the distributed algorithm.

How to Employ Reverse Search in Distributed Single Source Shortest Paths

by Lubo¹ Brim, Ivana Èerná, Pavel Krèál, Radek Pelánek, This is a full version of the paper presented at SOFSEM 2001. November 2001, 22 pages.

FIMU-RS-2001-09. Available as Postscript, PDF.

Abstract:

A distributed algorithm for the single-source shortest path problem for directed graphs with arbitrary edge lengths is proposed. The new algorithm is based on relaxations and uses reverse search for inspecting edges and thus avoids using any additional data structures. At the same time the algorithm uses a novel way to recognize a reachable negative-length cycle in the graph which facilitates the scalability of the algorithm.

Distributed Shortest Paths for Directed Graphs with Negative Edge Lengths

by Lubo¹ Brim, Ivana Èerná, Pavel Krèál, Radek Pelánek, This is a full version of the paper presented at FST&TCS 2001. May 2001, 19 pages.

FIMU-RS-2001-01. Available as Postscript, PDF.

Abstract:

A distributed algorithm for single source shortest path problem in an arbitrary directed graph which can contain negative length cycles is presented. The new algorithm is a label-correcting one and uses a novel way for detection of negative length cycles. It works on a network of processors with disjoint memory that communicate via message passing. Correctness of the algorithm is proved. The algorithm is work-efficient as its worst time complexity is O(n^3/p), where p is the number of processors.

Distributed LTL Model-Checking in SPIN

by Jiøí Barnat, Lubo¹ Brim, Jitka Støíbrná, December 2000, 19 pages.

FIMU-RS-2000-10. Available as Postscript, PDF.

Abstract:

Distributed version of the SPIN model checker has not been extended to allow distributed model-checking of LTL formulas. This paper explores the possibility of performing nested depth first search algorithm in distributed SPIN. A distributed version of the algorithm is presented, and its complexity is discussed.

A Fully Abstract Semantics for a Version of Synchronous Concurrent Constraint Programming

by Jean-Marie Jacquet, Lubo¹ Brim, David Gilbert, Mojmír Køetínský, December 1999, 62 pages.

FIMU-RS-99-08. Available as Postscript, PDF.

Abstract:

Concurrent constraint programming is classically based on asynchronous communication via a shared store. In previous work, we presented a new version of the ask and tell primitives which features synchronicity, our approach being based on the idea of telling new information just in the case that a concurrently running process is asking for it. We turn in this paper to a semantic study of this new framework, called Scc.

It is first shown to be different in nature from classical concurrent constraint programming and from CCS, a classical reference in traditional concurrency theory. This suggests the interest of new semantics for Scc. To that end, an operational semantics reporting the steps of the computations is presented. A denotational semantics is then proposed. It uses monotonic sequences of labelled pairs of input-output states, possibly containing gaps, and ending - according to the logic programming tradition - with marks reporting success or failure. This denotational semantics is proved to be correct with respect to the operational semantics as well as fully abstract.