Abderezak Touzene

A multilevel partitioning approach for efficient tasks allocation in heterogeneous distributed systems

This work addresses the problem of allocating parallel application tasks to heterogeneous distributed computing resources, such as multiclusters or Grid environments. The proposed allocation scheme is based on a multilevel graph partitioning and mapping approach. The objective is to find an efficient allocation that minimizes the application completion time, subject to the specified constraints pertinent to the application and system environment. The allocation scheme consists of three phases; the clustering phase, the initial mapping phase and the refinement and remapping phase. The scheme introduces an efficient heuristic in the clustering phase for contracting (coarsening) large size application graphs to the number of processors, called the VHEM method. An initial mapping technique based on a tabu-search approach has been introduced as a basis for the process of refinement and remapping phase. The simulation study shows that the VHEM coarsening heuristic can achieve optimal or near-optimal communication, compared to the HEM method, when the ratio of the number of tasks to the number of processors exceeds a threshold value. The simulation study shows that those optimal or near-optimal VHEM-coarsened graphs have an effect of generating very efficient mappings, when they are compared to the HEM-coarsened graphs.

Optimal all-ports collective communication algorithms for the k -ary n -cube interconnection networks

The need for collective communication procedures such as One-to-All broadcast, All-to-All broadcast arises in many parallel or distributed applications. Many of these communication procedures have been studied for many topologies of interconnection networks such as hypercubes, meshes, De Bruijn, star graphs and butterflies. In this paper we propose a construction of multiple edge-disjoint spanning trees for the k-ary n-cube which can be used to derive an optimal and fault tolerant broadcasting algorithm. We propose also an optimal All-to-All broadcasting algorithm. We consider the k-ary n-cube as a point-to-point interconnection, using store-and-forward, all-port assumption and a linear communication model.

On All-to-All Broadcast in Dense Gaussian Network On-Chip

Gaussian networks are gaining popularity as good candidates Network On-Chip (NoC) for interconnecting Multiprocessor System-on-Chips (MPSoCs). They showed better topological properties compared to the 2D torus networks with the same number of nodes N and the same degree 4. All-to-all broadcast is a collective communication algorithm used frequently in many parallel applications. Recently, Z. Zhang et al. [1] have proposed an all-to-all broadcast algorithm for Gaussian on-chip networks that achieves the minimum delay time but requires 4k extra buffers per router, where k is the network diameter. In this paper, we propose a new all-to-all broadcast algorithm for dense Gaussian on-chip networks that achieves the minimum delay time without requiring any extra buffers per router. In this paper, we propose a new all-to-all broadcast algorithm for dense Gaussian on-chip networks that achieves the minimum delay time without requiring any extra buffers per router. Along with low latency, reducing the amount of buffer space and power consumption are very important issues in NoCs architectures.

Performance of algebraic graphs based stream-ciphers using large finite fields

Algebraic graphs D(n, q) and their analog graphs D(n, K), where K is a finite commutative ring were used successfully in Coding Theory (as Tanner graphs for the construction of LDPC codes and turbo-codes) and in Cryptography (stream-ciphers, public-keys and tools for the key-exchange protocols. Many properties of cryptography algorithms largely depend on the choice of finite field Fq or commutative ring K. For practical implementations the most convenient fields are F and rings modulo Z modulo 2m. In this paper the reader can find the first results about the comparison of D(n, 2m) based stream-ciphers for m = 8, 16, 32 implemented in C++. They show that performance (speed) of algorithms gets better when m is increased.

A Tensor Sum Preconditioner for Stochastic Automata Networks

Stochastic automata networks (SANs) have gained great interest within the performance-modeling community because of their natural ability to capture parallel and distributed activities. SANs and the related concept of stochastic process algebra are advantageous because they keep the transition matrix in a compact form called the SAN descriptor. Several iterative and projection methods have been tested for SANs. Some preconditioners for SANs have been developed to speed up the convergence. Recently, Langville and Stewart [Langville, A., W. Stewart. 2004. A Kronecker product approximate preconditioner for SANs. Numer. Linear Algebra Appl.11 723--752] proposed the nearest Kronecker product (NKP) preconditioner for SANs with great success. Encouraged by their work, we propose a new preconditioning method, called the tensor sum preconditioner (TSP), which uses a tensor sum preconditioner rather than a Kronecker product preconditioner. In TSP, we take into account as much as possible the effect of synchronization using term grouping, factorizations, and approximation techniques. We conducted an experimental study to compare our TSP with the NKP preconditioner, and the results show that TSP outperformed NKP on both the number of iterations and the execution time.

Edge-disjoint spanning trees for the generalized butterfly networks and their applications

The generalized butterfly GBN(d,n) has recently gained some interest as a point-to-point interconnection network rather than the well known multi-stage butterfly networks. We construct edges-disjoint spanning trees (abbreviated EDSTs) for the GBN(d,n). Our construction is based on the decomposition of the GBN(d,n) into d^n vertex-disjoint cycles of length n. As an application, we propose an efficient broadcasting algorithm and its fault-tolerant version for the GBN(d,n). Our fault-tolerant broadcasting algorithm is optimal in terms of fault-tolerance, because it resists 2d-1 edge failures (2d is the degree of the GBN(d,n)). We also propose an efficient scattering algorithm and its fault-tolerant version which resists 2d-3 edge faults.

PARALLEL ROUTING IN MOBILE AD-HOC NETWORKS

This paper proposes and evaluates a new position-based Parallel Routing Protocol (PRP) for simultaneously routing multiple data packets over disjoint paths in a mobile ad-hoc network (MANET) for higher reliability and reduced communication delays. PRP views the geographical region where the MANET is located as a virtual 2-dimensional grid of cells. Cell-disjoint (parallel) paths between grid cells are constructed and used for building pre-computed routing tables. A single gateway node in each grid cell handles routing through that grid cell reducing routing overheads. Each node maintains updated information about its own location in the virtual grid using GPS. Nodes also keep track of the location of other nodes using a new proposed cell-based broadcasting algorithm. Nodes exchange energy level information with neighbors allowing energy-aware selection of the gateway nodes. Performance evaluation results have been derived showing the attractiveness of the proposed parallel routing protocol from different respects including low communication delays, high packet delivery ratios, high routing path stability, and low routing overheads.

Edges-disjoint spanning trees on the binary wrapped butterfly network with applications to fault tolerance

In many parallel applications, the need for broadcasting, scattering, gathering or gossiping is crucial. Many collective communication algorithms have been studied for different topologies of interconnection networks such as hypercubes, meshes, De Bruijn and star graphs. In this paper we study some communication procedures on the binary wrapped butterfly BWB(n) of dimension n interconnection networks. We consider the BWB(n) as a point-to-point interconnection network. Communication is assumed to be full duplex, all-ports with a linear communication model and is based on store-and-forward techniques. The BWB(n) is a constant degree 4 Cayley graph. Vadapalli and Srimani gave a new representation of the BWB(n) that bring some convenience in studying the topological properties and fault tolerance. Using this new representation, we propose an improved one-to-all broadcast algorithm, based on a spanning tree of optimal height. We present a technique based on rotative trees for constructing multiple spanning trees that would be used to derive: a fault tolerant one-to-all broadcast, a scattering, a gathering algorithms and theirs fault tolerant version.

Performance analysis of an extended grid based broadcast algorithm in mobile ad-hoc networks

In this paper we proposed a new extended grid-based broadcasting algorithm (EGBB) in mobile ad-hoc networks (MANETs) which reduces considerably the broadcast storm problem. EGBB algorithm is based on a logical 2-dimensional grid cells view of the geographical region of the MANET. EGBB algorithm minimizes the number of rebroadcasts by using gateway nodes for each grid cell, where only gateway nodes are responsible for rebroadcasting the message in order to solve network congestion problem. We extended the neighborhood of a node in a grid cell from the nodes located in only eight adjacent grid cells as originally in grid based broadcast (GBB) to any node in any grid cell within the node’s transmission range to increase the reachability and reduce the number of hops. We also added a new adaptive feature to suite different traffic load and mobility conditions. We developed a simulation model based on NS2 simulator to measure the performance of EGBB and compare the results with the GBB; the efficient counter based scheme (ECB); and the position-aware counter-based algorithm (PCB). The simulation experiments showed good results for our new algorithm EGBB. EGBB outperforms GBB, ECB and PCB in terms of end-to-end delay, number of saved rebroadcasts, and packet collision ratio under different traffic load, network density, and mobility conditions.

All-to-all broadcasting in torus Network on Chip

This paper proposes and evaluates the performance of an all-to-all broadcasting algorithm for a 2D torus Network on Chip (NoC). The proposed algorithm uses special spanning trees called NEWS spanning trees. These trees are link conflict free which implies that the communication steps of the all-to-all algorithm are contention free. The proposed all-to-all broadcasting algorithm is optimal in terms of transmission time and does not need any additional buffer memory like in the all-to-all algorithm for the 2D torus (IEEE Trans Comput 50:1029–1032, 2001). Reducing the amount of buffer space is a very important issue in NoC architectures. Our algorithm is therefore a more efficient solution for all-to-all broadcasting in 2D torus NoC multi-core systems compared to previously proposed algorithms.