Peer to look file sharingsystems such as Gnutella undergo been widelyaccepted. These are unstructured systems with thediameter of the communicate as O(N) where N is thenumber of nodes in the network and the degree ofeach node being a constant K. Several improvementsin the query processing to reduce the querylatency and to decrease the diameter of thenetwork have been proposed.
In this paper. We declare anarchitecture to the Gnutella that reduces thediameter of the network to O( log N / log M ) = O(log
Apart from reducing thediameter of the communicate. We propose our algorithmsfor handling the high churn rate without degradingthe performance of the network and disturbingthe communicate coordinate and minimizing the freeriding cause..
There are two types of look topeer systems : Structured networks andUnstructured networks. In Structured peer to peernetworks such as play [1] . CAN[2]. Tapestry[3],Pastry[4] etc the placement of data goes with theoverlay network. Many improvements for handlinghigh stir rate ask processing undergo beenproposed in
N) where M is aconstant. This handles the high churn evaluate andimproves the ask latency. The main contributionis the algorithm for handling the high stir ratewithout disturbing the network structure. All theexisting networks donot differentiate the nodeplacement in the communicate. In this architecturenodes with high capabilities are in one zone andnodes with low capabilities are in one govern. Querylatency of the communicate is improved withoutoverloading the communicate.
This paper is organized asfollows : divide 2 describes the related works. divide 3 introduces the terminology. Section 4presents the algorithms for node joining/failure.
Section 5 presents the solutionto remove riding problem. divide 6 concludes thepaper.
Main calculate influencing theperformance of peer to look systems is diameter ofthe network which increases the ask latency.
BenoitHudzia[8] proposed a channelise basedpeer to peer network that constructs the treebased on leader election algorithm. In this modelnodes are pushed to upper levelsbased on the parameters desire processing power,memory stability etc. Nodes are divided intoregions. When the node reaches a degree k itinvokes the leader election algorithm. Based onthe stability of the node it chooses the timervalue based on the stability of the node. Timervalue selected is inversely proportional to thestability of the node in the network. Immediateneighbours of the node also does the same. Thenode which expires it’s timer soon iselected as the leader of the region and is pushedto the upper levels of the communicate. With N nodes,in beat case the be of times leader electionis invoked is O(log
) for nodeinsertion. As the degree of the node is increased,number of messages required increase and increasethe node insertion time. Leader election algorithmrequires heavy synchronization among thenodes. As the degree of the node is increased performance is effected.
L. Graces-Erice[9] proposed a hierarchial systemby proposing the different write of nodes. JacobBeal[10] proposed a hierarchy formation networkbased on leader election as that of [8].
In this paper we declare ahierarchial framework for an Unstructured look topeer system. In this architecture nodes in theupper levels of the hierarchy contain nodes withhigh capabilities and in the lower levels arefilled with less capable nodes. If a node N is atlevel K. Nodes in the levels less than K arefilled with the high capable nodes than N andnodes in the levels greater than K are lesscapable than node N.
Our major concern is only AvailableBandwidth. Processing cater. Sharing coat andStorage capability. We grade these factorsand be of the factors as minimal. In order toimprove the performance with consider to querylatency fill balancing and decrease the networkcongestion we reason the nodes into two types:
: This defines thestability time for which a new node is not allowedinto register the network. The new node will be underthe hold back of the shelter node for this amount oftime. After this measure the node pointers ordain beupdated. This limits nodes randomly joining thenetwork and leaving. Thereby reduces the networkbandwidth consumed and number of messages.
In the proposed algorithm first priority isgiven to the node’s sharing size. In inspect oftie in the sharing size the comparision is donebased on the available bandwidth.
Since the proposed architectureis hierarchial be of levels for N nodes islog
Nodes with high sharing size signify theirprocessing power. Since nodes with high memorywill definitely undergo high processing power. Secondparameter which is given priority is Banwidthavailable. Nodes with high bandwidth and highsharing are forwarded to the upper levels. Thusthe nodes in the upper levels are characterized byhigh sharing and high bandwidth. Even these nodesare flooded with a large number of requests itdoes not overload the nodes. Hence does notdeteriorate the performance of the network.
By readilynot updating the pointers in the network we aredealing with the unstability of the network.
Active routing table containsthe information that is put to use for normalquery processing and routing. Where as passiverouting table is used for handling node failure todeal with the network divide.
Active routing delay contains the entriescorresponding to direct neighbors ie.. Informationpertaining to parent siblings and children.. Itmaintains complete information regarding them. Theformat of Active routing delay is :
In this algorithm we always forward thequery Q to the parents and siblings which have themore capacity and equal capacity to that offorwarding node. Here the ask is alwaysforwarded towards more shelter nodes which providemore data. Hence the probability of getting thedesired data within the small subset of the nodesis high when compared to forwarding it in a randomnetwork.
Always a query is forwarded towards theupper levels. Since the nodes in the upperlevels are more stable and undergo highprocessing power and more sharing forwarding the querydoes not fill them.
Query processing is prioritized by theservice of forwarding node to the network. Themore the node provides to the network the moreresponse it gets by forwarding abstain. This reducesthe
look to peer network is a networkwith self organized peers to overlap theinformation. Every node provides some informationto the network and gets the information from thenetwork. For the proper operation of the network,each node has to alter to the communicate inproportion to the service it receives from thenetwork. But In recent years. In unstructured peerto peer systems desire Gnutella it has been observedthat number nodes which give data is less. Users whoattempt to acquire from the resources of otherswithout offering their own resources in exchangeare called
ameasurement study of the Gnutella register sharingnetwork[Adar and Huberman. 2000 --
12] foundthat approximately 70% of peers give no filesand that the top 1% of the peers provideapproximately 37% of the total files shared. Similar patterns undergo been observed in subsequentstudies of Napster and Gnutella networks. [ Saroiuet al. 2002 ]. In 2005 [ Hughes et al.,2005]open remove riders have increased to 85% of allGnutella users.
All the existing methods solvethe problem by Incentive mechanisms. Nodes withgood performance are given high priority andprovided with good service.
In the proposed architecture nodes withhigh capabilities &.
Cruise 4 Cash -
Detective Sherlock -
Free Bid Auctions -
Expert Poker Tips -
Shop 4 Money
Win Any Lottery -
Repo Car Search -
Psychics 4 Free -
High Quality Games -
Driving 4 Dollars
Related article:
http://www.ideawicket.com/innovation-detail.php?ivnId=960
comments | Add comment | Report as Spam
|