Prof., Dr. Jarkko Vuori	Jarkko Vuori, Annemari Auvinen & Mikko Vapa

InBCT 3.2.

Peer-to-Peer communication/computation in M2M-environment (P2P)

Team Leader:

Prof., Dr. Jarkko Vuori,
Department of Mathematical Information Technology, University of Jyväskylä,
mimic@mit.jyu.fi, 014-2602750, 040-5012564

Background

Distributed systems have traditionally been based on the client-server architecture, in which centralized servers are used in communication. This solution is unsuitable for systems in which fault tolerance and load balancing play a prominent part, because server performance usually forms the bottleneck in these kinds of systems. Also, networks with rapidly changing topologies are generally difficult to manage in a centralized manner. Peer-to-Peer (P2P) networks (Napster and Gnutella, for example, for the distribution of multimedia files) are now challenging the client-server solution. They represent a fundamental change in distributed systems architecture. Each node in the P2P network acts both as a server and the client. As a result, data storage and processing can be distributed, and the bottleneck found in the client-server architecture can be eliminated.

The Cheese Factory project investigates the distribution of computational processing to many equal processing units. This is known as Peer-to-Peer processing. P2P processing does not employ any Master unit to manage the system: supervisory functions (task sharing, for example) are also distributed. The advantages of this solution are fault tolerance and ease of maintenance and supervision. Its disadvantages are complexity of management algorithms and greater need of communication resources.

Project Status

During the first project year period we defined and implemented the basic operations of the Chedar distributed P2P-platform. We have also implemented control and management software (P2P Studio) for the P2P-network based on our Chedar platform. P2P Studio can control the connections of a computation node and measure various system parameters from a node.

During the project year 2003 we will place a data fusion application on top of the Chedar distributed platform. We will also improve the resource locating services of the Chedar platform. Data fusion was selected for a test application, because it's processing and data acquisition requirements are relatively demanding. The Data fusion application was developed mainly for sensor network data gathering.

The project results for the year 2003 will consist of:

• An improved version of the Chedar distribution platform (Java component library)
• An improved Chedar network management software (P2P Studio)
• Publication: Weaknesses of centralized topology in Grid Computing environments
• Master's Thesis: A P2P distribution platform using JXTA platform

Future

Once the data fusion prototype is completed, we will investigate the behavior of Chedar's adaptive search algorithms in a malfunctioning workstation network and will publish a system level description of the distributed platform. Due to some faults in Sun's JXTA platform, we have commenced an experimental implementation of a streamlined JXTA substitute. This enables us to develop our own lower level data communication algorithms (search operation, for example) and compare them with JXTA. Another aim is to investigate scalability with the help of the Network Simulator 2, which we anticipate will give us new know-how for the development of Chedar.

Potential Applications

As embedded intelligence is becoming more and more common with everyday devices due to the lowered costs of component technology, the need for networking these devices increases (for example, the mobile phone or a voting procedure can be used to control house temperature). Typically, these networks have several mobile components (a mobile phone, car, etc.). P2P networks are very practical in many respects for the management of these types of tasks - they can make resource lookup and communication between the participating peers more efficient.

For example, if the aim is to determine the room temperature by voting without divulging the participants' preferences, P2P processing is an excellent way to do this. A mobile phone sends its temperature preference encrypted by a public key to another mobile phone, which sums them up (within certain limits, messages protected by a public key can be summed up without de-encrypting them). Once all the temperature preferences have been added up using the P2P network formed by the mobile phones, the room temperature control receives the sum total and can, with its key, average the sum, hiding all the private preferences. With this P2P network based method all kinds of polls can be conducted without divulging private opinions.

InBCT 3.2

Peer-to-Peer communication/computation in M2M-environment (P2P)

Tässä hankkeessa tutkitaan laskennan hajauttamista useille, tasa-arvoisille laskentayksiköille. Tämäntyyppistä laskentaa kutsutaan vertaisverkkolaskennaksi. Vertaisverkkolaskennassa ei ole lainkaan Master-yksikköä joka huolehtisi järjestelmän toiminnasta, vaan myös kaikki ylläpitoon tarvittavat toiminnallisuudetkin (esim. tehtävien jako) on hajautettu. Ratkaisun etuna on vikasietoisuus, skaalautuvuus ja ylläpidon helppous. Haittana on taas hallinta-algoritmien monimutkaisuus ja suurempi resurssien tarve. Vertaisverkkolaskentaan läheisesti liittyvät agentti-suuntautuneet menetelmät.

Hankkeessa modifioidaan keskitetty Globus laskennan hajautusympäristö siten että siitä saadaan aidosti vertaisverkkolaskentaan soveltuva. Muutos toteutetaan lisäämällä korvaamalla keskitetty resurssien hallintamoduuli omalla hajautetulla CheDAr (Acheap Distributed Architecture) vertaisverkkomoduulilla. CheDAr toteutetaan JXTA-alustan päälle käyttäen Java-ohjelmointikieltä. Ratkaisun toimivuutta evaluoidaan kehittämällä sensoriverkon datafuusiosovellus käyttäen CheDAr-alustaa.

Lisätietoja yms.:
More Information etc.:

Jarkko Vuori: InBCT 3.2 - Selvitys vuodelta 2003 (DOC, 35 KB).

Jarkko Vuori: InBCT 3.2 Peer-to-Peer communication/computation Cheese Factory -project (PPT, 798.5 KB) (Introduction in Finnish)

Jarkko Vuori: InBCT 3.2 Peer-to-Peer communication/computation Cheese Factory -project (PPT, 839 KB) (Introduction in English)

Sergiy Nazarko: Evaluation of Data Fusion Methods using Kalman Filtering and Transferable Belief Model (PDF, 1.6 MB), Master's Thesis, 28.11.2002

Sergiy Nazarko: Distributed Data Fusion in Peer-to-Peer Environment (PPT, 458 KB)

Vira Smirnova: Multi-Agent System for Distributed Data Fusion in Peer-to-Peer Environment (PDF, 898.1 KB), Master's Thesis, 28.11.2002.

Mikko Vapa & Matthieu Weber: Chedar - User Stories (PDF, 113.1 KB)

Mikko Vapa: Power-Law Networks (PPT, 743 KB)

Matthieu Weber: Chedar - Peer-to-Peer Platform (PDF, 53.4 KB) (Seminar 13.3.2003)

Kotilainen N., Introduction to Peer-to-Peer Studio (PPT, 103 KB).

Jarkko Vuori: P2P (PPT, 663.5 KB)

Dr. Jarkko Vuori

Cheese Factory

datafuusio.ppt (PPT, 34 KB)

Annemari Auvinen. Topologian hallinta -algoritmit Chedar-vertaisverkkoalustassa (PDF, 1.7 MB). Vertaisverkot sopivat hyvin tiedon levittämiseen hajautetusti, mutta niiden ongelmana on, miten solmun yhteyksiä lisätään ja poistetaan dynaamisesti muotoutuvassa verkossa ja miten resurssien haku saadaan mahdollisimman tehokkaaksi. Tämän tutkielman teoriaosuudessa käsitellään vertaisverkkoja ja graafeja sekä aiheeseen liittyviä aikaisempia tutkimuksia. Käytännön osassa tutkitaan Chedar-vertaisverkkoalustaan kehitettyjen topologian hallinta -algoritmien toimintaa keskittyen niiden muodostamiin topologioihin.

Annemari Auvinen ja Mikko Vapa. Minimaaliset Steiner-puut ja vertaisverkot (DOC, 72.5 KB). Dokumentissa kuvataan mitä ovat minimaaliset Steiner-puut ja kuinka niitä voidaan soveltaa optimaaliseen resurssin hakuun vertaisverkoissa.

Yevgeniy Ivanchenko. Using MatLab Self-Organizing Map (SOM) toolbox for investigation of NeuroSearch (DOC, 1014.5 KB). The assignment represents results of investigation of statistical data obtained from NeuroSearch (trained with evolutionary algorithms). As analyzing tool Self-Organazing Map (SOM) toolbox for MatLab was used. Also the assignment briefly describes method for processing of data obtained from simulation using Breadth-First Search (BFS) algorithm. Results of this processing will be used later for training of Multi-Layer Perceptron (MLP).

Mikko Vapa, Niko Kotilainen, Heikki Kainulainen and Jarkko Vuori. Resource Discovery in P2P Networks Using Evolutionary Neural Networks (PDF, 332.1 KB). Resource discovery is an essential problem in peer-to-peer net-works since there is no centralized index where to look for infor-mation about resources. One solution for the problem is to use a search algorithm that locates resources based on the local knowl-edge about the network. Traditionally, these search algorithms have been based on few predetermined rules. The problem with these algorithms is that if the conditions in the network change the algorithm becomes less efficient and won\'t adapt to the new envi-ronment. In this paper, we describe the results of a process where evolutionary neural networks are used for finding an efficient search algorithm. The initial test results indicate that an evolution-ary optimization process can produce search algorithm candidates that are more efficient compared to breadth-first search algorithm (BFS) used in Gnutella peer-to-peer network.

Juha Keto-Tokoi, Mikko Vapa ja Teemu Vesalainen. Polttoainehinta -mobiilipalvelun tutkimus (DOC, 3.3 MB). Polttoainehinta -mobiilipalvelun avulla käyttäjät saavat tietoon lähimpien bensa-asemien hintatiedot ilmaisesti vertaisverkon välityksellä. Tutkimuksessa selvitettiin kuinka tärkeäksi ihmiset kokevat Polttoainehinta -mobiilipalvelun ja paljonko he ovat siitä valmiita maksamaan. Lisäksi tutkimuksessa löydettiin käyttäjäsegmenttejä, jotka voisivat palvelusta hyötyä.

Annemari Auvinen, Niko Kotilainen and Mikko Vapa. Parallelization of Evolutionary Algorithms in Peer-to-Peer Environment (PDF, 179.5 KB). The document introduces how evolutionary algorithms/genetic algorithms can be parallelized for multiple workstations using Chedar peer-to-peer platform for distribution.

Matthieu Weber. Peer-to-Peer Replication (PDF, 220.3 KB). The document introduces the replication methods used in peer-to-peer networks.

Mikko Vapa, Niko Kotilainen, Annemari Auvinen, Joni Töyrylä, Hermanni Hyytiälä and Jarkko Vuori. NeuroSearch: evolutionary neural network resource discovery algorithm for peer-to-peer networks (PDF, 694 KB). Discovery of distributed resources is an essential problem in peer-to-peer (P2P) networks since there is no centralized index where to look for information about resources. One solution for the problem is to use a search algorithm that locates resources based on the local information about the network. Traditionally, the search algorithms have been designed to be based on few rules and designed completely by humans. The problem with these algorithms is that if the conditions like resource distribution and topology in the network change the algorithm becomes less efficient and won\'t have the flexibility to adapt to the new environment.

In this paper we describe the results of a process where evolutionary neural networks are used for finding an efficient search algorithm. By using this process we are able to define the network conditions and the quality of the search algorithm beforehand as an optimization problem and solve it computationally. As an end product we get a search algorithm that has adapted to the predefined conditions. The test results indicate that the approach is feasible and that an evolutionary optimization process can produce candidates that are competent compared to breadth-first search algorithm used in peer-to-peer networks.

Vira Smirnova. Multi-Agent System For Distributed Data Fusion In Peer-to-Peer Environment (PDF, 898.1 KB). This thesis work was aimed to design a new model for fault-tolerant and high availability of Distributed Sensor Networks and data fusion. The main problem was to modify existing data fusion process into a new one by combining multi-agent systems with P2P environment. In the research process was developed Peer-to-Peer distributed Sensor Network model and described a system using components and agents with several scenarios and states. The proposed design also considers the behaviour of the system in common failure conditions. Finally a comparison between DSN, MADSN and P2PDSN was formed using general metrics used for analysing complex distributed systems.

Olli Alanen, Kimmo Haukimäki, Timo Juonoja and Petri Rönkkö. MoPeDi Project - Project Report (PDF, 301 KB). MoPeDi project developed a generic middleware component that is used between Bluetooth\'s transmission protocols and different kind of mobile peer-to-peer applications. The product is called BlueCheese. The test application in the project was a dummy application with needed features. The project report describes how the project was carried out and compares the realized project to the project plan.

Kimmo Haukimäki. Mobiilien vertaisverkkojen väliohjelmistot (PDF, 454.4 KB). Tutkielma johdattelee tämän hetkiseen mobiiliin vertaisverkko maailmaan. Siinä määritellään keskeisimmät termit aiheeseen liittyen sekä tutkitaan yhteensä viiden mobiilin vertaisverkko väliohjelmiston rakennetta ja ominaisuuksia.

Sergiy Nazarko. Evaluation of Data Fusion Methods Using Kalman Filtering And Transferable Belief Model (PDF, 1.6 MB). In the research work data fusion algorithms, which can be used for target identification and target tracking were evaluated. The aim of this work was to provide a clear picture about data fusion, particularly in case of target tracking. For doing this tracking tool application was developed as research environment, containing Kalman filtering algorithm and transferable belief model implementations. The research process was concluded by validating the implementations using simulation settings found from research literature. Analysis showed the same performance as other research results and the evaluation was found successful.

Tiina Onkila, Hannakaisa Raiskinmäki, Mikko Vapa ja Teemu Vesalainen. Agora Centerin InBCT-hankkeen markkinointisuunnitelma (DOC, 150.5 KB). Tutkimustyömme pureutuu Jyväskylän yliopiston Agora Centerin Innovations in Business, Communication and Technology (InBCT) -hankkeeseen. Hanke on poikkitieteellinen ja koostuu useista tutkimusprojekteista, jotka liittyvät ihmislähtöisen teknologian tutkimukseen ja kehittämiseen. Työn tavoitteena on kuvata mahdollisimman hyvin mitä tekijöitä hankkeeseen liittyy sekä esittää joitakin parannuksia siihen miten InBCT-hanketta voitaisiin tehokkaammin markkinoida yksityisen ja julkisen sektorin asiakkaille. Tutkimustyö tehtiin kevään 2004 aikana taloustieteiden tiedekunnan Markkinoinnin suunnittelu -kurssille. Lähdemateriaalina käytimme henkilökohtaista haastattelua InBCT-hanketta tuntevan työntekijän kanssa sekä virallisia Agora Centeristä ja sen toiminnasta kertovia esitteitä ja WWW-sivuja.

Niko Kotilainen, Ville Pentti, Jussi Rastas and Joni Töyrylä. Guardian Project - Project Report (PDF, 371.8 KB). Guardian project implemented a testing and monitoring tool - P2PStudio - for a Chedar network developed by Agora Center\'s InBCT/Cheese Factory Project. This document describes how the project was carried out.

Sergiy Nazarko, Mikko Vapa, Annemari Auvinen and Jarkko Vuori. Decentralized Data Fusion In Peer-to-Peer Environment (DOC, 271 KB). This paper describes decentralized data fusion system in peer-to-peer environment. Decentralization plays crucial role in military services. If there is no central in the system then the risk of critical failure is minimized. Nowadays peer-to-peer networks and services are one of the most evolving technologies. P2P sensor networks with data fusion capabilities built on mobile platform will become an up-to-date solution in applications of military defence. This article includes theoretical and practical aspects of decentralized data fusion system implementation in multi-sensor peer-to-peer environment. Two algorithms have been taken into consideration during the research. They are Kalman Filter, and Transferable Belief Model. In this article we will concentrate on the Kalman Filter.

Matthieu Weber. Broker-based Architecture for Advertisement Propagation in P2P Networks (PDF, 7.4 KB). This paper considers advertisements exchanged in peer-to-peer networks. The nodes searching for a given resource send purchase advertisements, and nodes sharing their resources with others send sale advertisements. Locating a resource on the network then consist in matching purchase advertisements with sale advertisements. The architecture described here is based on brokers, which are nodes specialized in matching advertisements.

Matthieu Weber, Jarkko Vuori and Mikko Vapa. Advertising Peer-to-Peer Networks over the Internet (PDF, 127.6 KB). Most of the peer-to-peer overlay networks either provide a centralized way to join the network or rely on out-of-band methods for that purpose. None of them is satisfactory, since the first one relies on a single point which can fail, thus making the network vulnerable to failure and the second one gives no solution to the problem. This document proposes a new approach to the problem, using already existing systems such as Usenet, IRC and Web search engines to advertise the presence of an overlay network on the Internet and thus facilitate the binding of new nodes into such a network, in a fault tolerant way.

Joni Töyrylä. Building NeuroSearch - Intelligent Evolutionary Search Algorithm For Peer-to-Peer Environment (PDF, 2.6 MB). Tutkimuksessa kuvataan evolutionaarisen hakualgoritmin kehittymisprosessi ja sen tehokas käyttö vertaisverkoissa. Tutkimuksessa pohditaan kuinka neuroverkon rakenteelliset muutokset, erilaiset sisääntuloarvot ja hyvyysfunktion muuttuminen vaikuttavat hakualgoritmin kehittymiseen ja toimintaan. Tutkimuksessa selvitetään minkälaisia eri luonteenpiirteitä neuroverkkohakualgoritmeilla voi esiintyä ja hahmotellaan rajoituksia, jotka tämän tyyppisessä hakualgoritmin suunnittelussa on olemassa.

Yevgeniy Ivanchenko. Adaptation of Neural Nets For Resource Discovery Problem in Dynamic and Distributed P2P Environment (PDF, 1.1 MB). Peer-to-Peer networks are complex systems that have widely spread recently and thus providing efficient management to them is necessary and at the same time challenging task. The aims of this master's thesis work are: discovering of decision mechanism of NeuroSearch algorithm developed in the Agora Center, testing NeuroSearch under dynamic conditions and finding good solution for optimization problem. These three goals are lying on the middle part of the path to solving the main problem: developing new approach to train the core mechanism of NeuroSearch, which is based on the MLP, under dynamic conditions and in supervised manner.