Kei Ohnishi

Dynamic storage load balancing with analogy to thermal diffusion for P2P file sharing

This paper presents a file replication scheme with analogy to thermal diffusion for storage load balancing in unstructured peer-to-peer (P2P) file sharing networks. The proposed scheme is intended to balance storage load among peers in a dynamic, distributed, and autonomous manner, as in thermal diffusion. Theoretical analysis results show that the presented scheme actually has a statistical analogy with a thermal diffusion equation. In addition, the proposed scheme includes parameters to widely explore the trade-off between storage load balancing and search performance existing in unstructured P2P file sharing networks. Simulation results show that compared to other replication schemes, the proposed scheme has better ability not only in balancing storage load among peers, which is the primary objective of the proposal, but also in widely exploring the performance trade-off.

Multi-Jain Fairness Index of Per-Entity Allocation Features for Fair and Efficient Allocation of Network Resources

Due to its simplicity and its easy comprehension, Jains fairness index is still among the most popular measures to compare justness of allocations. However, it was already argued in the original paper that while the way of computing the index is well established, it is not immediately clear to which metric to apply the computation. Thereby, metric stands for a specific choice of a system observable. Here we study the extension of Jains index to multiple metrics at once. We propose a set of per-entity allocation features to represent justness of an allocation, and to derive corresponding vectors of feature-wise taken Jains fairness indices. The features give a numerical representation of fulfilling common fairness properties like proportionality, envy-freeness and equity of an allocation. Then, maximizing the smallest index gives an efficient procedure for allocation of goods. We study this procedure for the problem of allocating wireless channels in a multi-user setup and compare the influence of the various feature choices on the efficiency of the solution.

Color Effect on Subjective Perception of Progress Bar Speed

This present study attempts to find relationship between progress bar colors and subjective speed by subjective evaluation experiment. We prepared the six combinations of colors, blue/red for progress bar foreground color and cyan/orange/gray for background color. The test progress bars are designed under the same condition, for example progress bar size or animation speed, except colors. The test progress bars were displayed one after the other of all pair combinations, and made the subjects mark the test progress bar which they felt faster. As a result, there were no obvious significant color effects related to subjective speed impression in this subjective evaluation experiments. We consider the reason why not found significant color effects as follows, (i) its not enough data due to the small-scale experiments with only 10 subjects, (ii) color effect might not strong on progress bar.

Meta-heuristic optimization reloaded

We consider the meta-heuristic approach to optimization as to be performed in four stages (model, optimality, algorithm, verification), and point out the potential of varying the optimality stage, in contrary to the design of new algorithms. Thus, we can also apply the meta-heuristic approach to optimization to the task of fair distribution of indivisible or elastic goods, where the optimality is represented by (set-theoretic) fairness relations. As a demonstration, we fix a meta-heuristic algorithm (here a generalized version of the Strength Pareto Evolutionary Algorithm SPEA2) and provide a set of 15 fairness relations, along with the discussion of general design principles for relations, to handle the Wireless Channel Allocation problem. For validation, comparison with an equal-effort random search is used. The demonstration shows that while all relations represent a similar model (they are all directly or indirectly related to the Bottleneck Flow Control algorithm), the performance varies widely. In particular, representing fairness of distribution by ordered proportional fairness or by exponential Ordered-Ordered Weighted Averaging appears to be in favour of a successfull meta-heuristic search.

A GRATIS theorem for relational optimization

We are studying the NFL Theorems with regard to relational optimization. In relational optimization, we represent the optimization problem by a formal relation, and the solution by the set of maximal (or non-dominated) elements of this relation. This appears to be a natural extension of standard optimization, and covers other notions of optimality as well. It will be shown that in this case, the NFL Theorems do not hold and that there are pairs of algorithms and a performance measure, where one always outperforms the other with regard to this performance measure if averaged over all possible relations. More specifically, the outperforming algorithm is an algorithm, where elements are checked one by one if they are dominated by any other element, while the outperformed algorithm is an algorithm, where elements are checked one by one if they dominate any other element. The proof is accompanied by a complete analysis of a special case, where also other performance measures are shown to be better when using the former algorithm.

Autonomously Reconstructable Semi-Structured P2P Networks for File Sharing

We propose a semi-structured peer-to-peer (P2P) network for file sharing built from a structured P2P network as a basis for file search together with an unstructured P2P network for access load balancing. This semi-structured P2P network adopts a structured P2P network capable of providing reliable file search as its base structure. In addition, it makes lightly loaded nodes in the structured network construct unstructured P2P networks over which heavily loaded nodes can spread their access load. The semi-structured P2P network is required to dynamically reconstruct its network topology according to changes in node loadings and file search performance. We also propose an autonomic, distributed, and cooperative method for dynamically altering the network topology. In simulation experiments, we first obtain a network topology that ensures reliable file search and is optimal in terms of access load balancing. This is carried out under the (idealized) assumption that the load state of all nodes in the network can be obtained in advance. We also show that the proposed topology reconstruction method yields network topologies with almost the same performance as the optimal configuration.

Maxmin Fairness under Priority for Network Resource Allocation Tasks

Many network control policies can benefit from introducing priorities among users, traffic flows, or service provisions e.g. For QoS improvement or network congestion avoidance. In order to ensure fairness of concomitant resource sharing tasks, generic extensions of maxmin fairness under priority are considered. A critical analysis of existing approaches leads to the definition of two fairness relations based on formal modifications of the maxmin fairness standard of comparison. One is based on using priority functions for internal weighting, the other on priority classes. Experimental evaluation shows that maximizing under these relations indeed gives relations close to maxmin fairness that take priorities into account such that higher priority users receive higher allocations in average. Further studying the wireless channel allocation model problem shows that prioritizing can give solutions of higher efficiency than maxmin fairness.

Meta-heuristic approach to proportional fairness

Proportional fairness is a concept from resource sharing tasks among n users, where each user receives at least 1/n of her or his total value of the infinitely divisible resource. Here we provide an approach to proportional fairness that allows its extension to discrete domains, as well as for the direct application of evolutionary computation to approximate proportional fair states. We employ the concept of relational optimization, where the optimization task becomes the finding of extreme elements of a binary relation, and define a proportional fairness relation correspondingly. By using a rank-ordered version of proportional fairness, the so-called ordered proportional fairness, we can improve the active finding of maximal proportional fair elements by evolutionary meta-heuristic algorithms. This is demonstrated by using modified versions of the strength pareto evolutionary algorithm (version 2, SPEA2) and multi-objective particle swarm optimization. In comparison between proportional and ordered proportional fairness, and by using relational SPEA2, the evolved maximum sets of ordered proportional fairness achieve 10 % more dominance cases against a set of random vectors than proportional fairness.

Parallel Evolutionary P2P Networking for Realizing Adaptive Large-Scale Networks

The present paper proposes a parallel evolutionary P2P networking technique (P-EP2P) that applies an evolutionary networking technique (EP2P) in parallel to small networks into which an entire P2P network is divided. EP2P dynamically and adaptively optimizes several P2P network topologies, in which all of the nodes are included at the same time, in an evolutionary manner. However, EP2P has a scalability problem that load is more concentrated in a particular node as the number of nodes increases. The aim of P-EP2P is to first balance load among the nodes to maintain the function of the entire network and then bring adaptability even to a large-scale P2P network. Simulation results reveal that P-EP2P is capable of balancing load among the nodes, but degrades search reliability as the number of the small networks increases, which indicates that there is a trade-off relationship between the load balancing and the search reliability. The results suggest that to enhance the search reliability, the small networks should exchange information among them to maintain consistency of the network topologies, though that causes additional load to the nodes.

Thermal Diffusion-Based Access Load Balancing for P2P File Sharing Networks

In the present paper, we propose a file replication method to dynamically balance loads of peers in unstructured peer-to-peer (P2P) file-sharing networks. Load balancing is considered to require uniform storage access by file replication. We then define the load index as the storage access ratio. In the proposed method, each peer autonomously regulates the file replication probability to uniform storage accesses as thermal diffusion phenomena. Simulation results reveal that the proposed method can control the file sharing network to balance loads while dynamically adapting to change of storage accesses that is caused by change of popularity trends, and therefore keep load balancing performance stable. In addition, the proposed method achieves greater adaptability to sudden changes in storage accesses than our previous method.