Published on

IJERA (International journal of Engineering Research and Applications) is International online, ... peer reviewed journal. For more detail or submit your article, please visit www.ijera.com

  • Be the first to comment

  • Be the first to like this

No Downloads
Total views
On SlideShare
From Embeds
Number of Embeds
Embeds 0
No embeds

No notes for slide


  1. 1. G. T. Chavan., M. A. Mahajan / International Journal of Engineering Research and Applications (IJERA) ISSN: 2248-9622 www.ijera.com Vol. 3, Issue 2, March -April 2013, pp.356-359 Load Balancing in P2P networks using DHT based systems and Ant based systems: A Comparison G. T. Chavan, M. A. Mahajan Asst. Prof. Sinhgad College of Engineering, Pune M.E. Student Sinhgad College of Engineering, PuneAbstract The heterogeneity in peer to peer (P2P) searched.networks can be its advantage as well as Distributed hash tables (DHTs) aredisadvantage. The heterogeneity may be about exampleamount of storage, computing power, of decentralized distributed systems. DHTsconnectivity etc. But P2P systems make it provide a lookup for the data items similar to apossible to harness resources such as the hashing where (key, value) pairs are stored in thestorage, bandwidth and computing power oflarge populations of networked computers in a DHT, and any node can efficiently get the valuecost effective manner. There are many associated with a given key. Responsibility of theapproaches to balance these systems in a cost- mapping for keys to values is divided among theeffective way. Here we are giving focus on two nodes, in such a way that a change in mappingmain approaches; the Distributed Hash Table causes a minimal amount of disruption. Figure 1(DHT) based systems and Ant based systems. shows the structure of the DHT where the dataBoth approaches are having its own advantages items are combined with the hash function and theand shortcomings. result generated will be stored in the node as a (key, Value) pair. DHTs form an infrastructure that can be used toKeywords—load balancing, p2p networks, build P2P networksDHT, Ant based Systems II. DHT WORKINGI. INTRODUCTION In the DHT based systems each data item which isPeer to Peer network:- to be stored on the computers is having its unique Peer to peer network is way of id. The identifier space is partitioned among theimplementing the networking between the nodes oncomputers where all are equivalent in nature i.e. allshare equivalent responsibilities of processing data. FOX HF DFA peer to peer network can also be created in ad-hoc connection; where a couple of computers CDconnected via a Universal Serial Bus to transfer FOX HF 345files, or it can also be a permanent network like a RUNS 52E 4network setup in a small organization.Peer to peer systems doesn’t have any centralised D87control or any hierarchy in operation, so peer to 9Epeer systems are distributed systems where the FOX HF WALKS 45Osoftware running at each node is equivalent infunctionality. 428 P 41 Fig.1. The structure of Distributed Hash Table. eLoad Balancing: Load balancing is sharing the work e the system, called as zone. Every node onbetween two or more computers, CPUs, network the system is responsible for storing all the items rlinks etc. The load balancing is done to maximize that are mapped to an identifier in its portion of the sthroughput- by getting the work done from all the space. The data items can be accessed and storedresources available in the system. As many on the zone by using a set of functions i.e. put(ID,resources are working simultaneously the resource item): stores the item associated with ID, andutilization will be higher and the response time will get(ID): retrieves the item corresponding to ID. Thegradually minimize. Load balancing can also be load balancing is done using the virtual servers.useful when dealing with redundant The virtual server is part of the node. A single nodecommunications links. To achieve balancing first can be partitioned into multiple virtual servers. Athe nodes should be available for processing and virtual server is also responsible for the identifierthe different routes between them should also be 356 | P a g e
  2. 2. G. T. Chavan., M. A. Mahajan / International Journal of Engineering Research and Applications (IJERA) ISSN: 2248-9622 www.ijera.com Vol. 3, Issue 2, March -April 2013, pp.356-359space. As the node is a combination of multiple conditions and selecting the destination for the loadvirtual servers it is responsible for multiple transfer. In semi-centralized the multiple centralidentifier spaces. The basic advantage of using nodes are responsible for the storing the loadvirtual servers is that when the node gets conditions and selecting the destination for the loadoverloaded at that time a virtual server can be transfer for the local systems. Although thetransferred to any other node making the source centralized system provides good load balancingnode lighter. This mechanism is supported by DHT but these system are immune to single pointby making two calls leave source node and join failures which leads to unbalanced system verydestination node. This gives advantage over the easily.scheme of having only one virtual server per node The decentralized systems will not affect form thewhere a node can transfer his load to its neighbours the single point failure problem but they sufferonly. from load information collection overhead. Due to the decentralized nature it needs more time forThere are multiple ways to do load balancing in the convergence.DHT based systems, some of which are explainedbelow. III. ANT COLONY OPTIMIZATION The first method is having one to one An Ant is a simple creature of nature. Antsmapping. In this scheme the light node will select always work in a group, usually called as colony.any ID randomly and will start searching for the Ants discover the shortest path to the food sourcenode who is responsible of that ID. When the target from its nest. For this ants use indirectnode is identified, its load is checked. If the target communication method called as stigmergy. Antsnode is heavy then the light node will initiate the achieve stigmergic communication by using atransfer operation and will accept the load of the chemical substance called as pheromone. Whileheavy node till its threshold value. The advantage searching for the path to the food source ants layof the scheme is that it is simple to implement; the down the pheromone on the path which is sensedlight node is doing probing not the heavy node so by the other ants. Ants will make the decisionthe heavy node is free from the Burdon of probing. depending upon the concentration of theThe second method is having the mapping of one to pheromone on the path; the more the concentrationmany. In this scheme one directory maintains the high is the probability to select the path. Computerlist of light nodes. The heavy node sends scientist transformed this model of collectiveinformation of its target load and the load of every intelligence into the computer algorithms. Thisvirtual server it has to the node which maintains the emerging field is called as Ant Colonydirectory. Optimization (ACO). In ACO analogous to the Then the directory node will search for the biological ants there is society of mobile agents.best possible light nodes that will accommodate the These mobile agents communicate depending uponvirtual servers of the heavy nodes and then will do the pheromone value which is nothing but thethe transfer accordingly. This procedure is repeated entries in the routing table made by the artificialtill all the heavy nodes will become light nodes. ants. The ACO can be applied to variousThe third scheme is many to many. In this scheme combinatorial problems like asymmetric travellingthere is pool of virtual servers, where every node salesman problem, graph coloring problem, vehiclewill submit the list of virtual servers to it. Then routing problem etc. To simulate the behaviour offrom the pool the heaviest virtual server is selected an ant ACO uses different data structures like toand transferred to the light node by making sure simulate laying and sensing of pheromone simplethat the destination node will not become heavy. If procedures are used. For example while doingthere is no node available to transfer the virtual migration from one node to another node, anserver then from the available list of light nodes, a artificial ant emulates laying of pheromone byvirtual server is selected and transferred to the pool updating the routing table entries in a node.and the heaviest virtual server is transferred to thenode.Disadvantages of the systems based on DHT are: The disadvantage of the systems based onthe DHT is that in these systems the processingrequired for the load collection and the loadtransferred are not considered. Some systems arecentralised and some of the systems aredecentralized. The centralised systems can furtherbe classified as full centralized and semi-centralized. In the full centralized systems one Fig. 2. Ant Colony Optimization.central node is responsible for storing the load 357 | P a g e
  3. 3. G. T. Chavan., M. A. Mahajan / International Journal of Engineering Research and Applications (IJERA) ISSN: 2248-9622 www.ijera.com Vol. 3, Issue 2, March -April 2013, pp.356-359Figure 2 is showing the typical working of ant node (the heavy node) and will destroy itself. Thecolony optimization. In this network sis nodes are source node now has the information of the lightpresent out of which one node acts as nest and one node and the shortest path to reach to the node; andnode acts as food source. Between the nest and will transfer part of the load to the light node. Infood source there are 4 more nodes present. There this way the load balancing is achieved. This hasare 4 named as A1, A2, A3, A4 ants initially advantages over the DHT based systemsforaging for food-seeking. There are multiple paths 1. The ACO is free from the single pointpresent between nest and food source; out of which failure due to which the centralized DHTtwo paths namely R1, R2 are shown in the figure. systems fails.A1, A2 randomly select the path R1 and A3, A4 2. ACO has the advantage over theselected randomly R2. While travelling through the decentralized systems also; that thepath ants will deposit pheromone on the path. decentralized systems suffer from the loadWhen A3 and A4 will reach the food source before information collection overhead. WhileA1 and A2 since R2 is having shorter length than applying ACO there is no need of loadR1. For their return way two paths are present R1 information collection.and R2, A3 and A4 will select R2 since thepheromone concentration on R2 is more than R1( IV. FUTURE SCOPEA1 and A2 are still not reached food source). The The ACO still is a part of the research. Therefurther ants will sence the pheromone are many areas where ant based solutions couldconcentration on the two paths and will select the work. The ACO also has some shortcomings like ifoath having more concentration. In this case it will a path is selected by all the ants then the pathbe R2 since it has more probability over R1. becomes stagnant, and the performance reducesWhat are the advantages to use ACO? gradually. So there was need to refine the ACO. 1) Routing Information The Multiple Ant Colony Optimization (MACO) iswhen the Ant Colony Optimization technique is used to overcome the shortcomings of the ACO.used the overhead of keeping the routing The MACO is same like ACO with littleinformation is reduced. Unlike in other networks to advancement. In MACO there are more onemaintain the routing table the needs the information colonies of ACO are working simultaneously. Eachfrom its neighbours to get the information of their of the ACO will find its own optimal path and atneighbours. This overhead is not present in the the source node the comparison between the set ofACO. In ACO each path is parallely identified and optimal paths will take place. Among the optimalthe routing table are also built randomly. paths the one which will give the shortest path will 2) Routing Overhead be selected. If the selected path becomes stagnantIn the networks when the routing table gets updated athen the other path from the pool of optimal pathsby flooding of link state update messages. For this is selected and the performance will not beevery node has to do the flooding after regular affected.interval. This overhead is removed in ACO whereonly pheromone tables are updated parallel. REFERENCES 3) Adaptively and Stagnation [1] Ananth R., Karthik L., et.al. “Loadthe network gets stagnant due to the flooded Balancing in Structured P2P Systems,”packets sent by all the nodes. In ACO only ant Proc. Second Int’l Workshop Peer-to-Peerpackets are sent to other nodes which are very Systems (IPTPS ’03), Feb. 2003small in size. [2] Brighten G., Karthik L., et.al., “Load Balancing in Dynamic Structured P2PHow ACO will do the load-balancing? Systems,” Proc. IEEE INFOCOM, 2004.The basic idea behind the load balancing is to [3] K. M. Sim, W. H. Sun., “Ant Colonytransfer the load of the heavy nodes and shift it the Optimization for Routing and Load-light nodes. In the DHT based system as explained Balancing: Survey and New Directions,”in the section II above; the nodes are searched by IEEE Trans. On Systems, Man, anddifferent techniques. In ACO the searching of the Cybernetics—Part A: Systems andlight nodes will be done by ants. As the ants forage Humans, Vol. 33, no. 5, Sept. 2003.for food; to do the load balancing they will search [4] Chyouhwa C. and Kun-Cheng T. “Thethe light nodes. While travelling the ant will reach server reassignment problem for loadto node they will check its load status and will balancing in structured P2P systems,”simultaneously update the pheromone table which IEEE Trans. Parallel and Distributedcontains the load value of a node. When the ant Systems, Vol. 19, Issue no. 2, Feb.will find the sufficient light node it will generate a 2008backward ant and destroy itself. The backward ant [5] Ion S., Robert M., et.al, “Chord: Awill follow the path depending upon the pheromone Scalable Peer-to-Peer Lookup Service forconcentration on it and will reach to the source Internet Applications,” Proc. ACM 358 | P a g e
  4. 4. G. T. Chavan., M. A. Mahajan / International Journal of Engineering Research and Applications (IJERA) ISSN: 2248-9622 www.ijera.com Vol. 3, Issue 2, March -April 2013, pp.356-359 SIGCOMM ’01, pp. 149-160, 2001.[6] Jonathan L. and Margo S., “Distributed, Secure Load Balancing with Skew, Heterogeneity, and Churn,” Proc. IEEE INFOCOM, 2005.[7] P.Brighten G., and Ion S., “Heterogeneity and Load Balance in Distributed Hash Tables,” Proc. IEEE INFOCOM, 2005.[8] Ziyong X. and Laxmi B., “Effective load balancing in P2P systems,” Proc. Sixth IEEE symposium Cluster computing and grid, Feb. 2006 359 | P a g e