A Grid Resource Discovery Model Based On Layered Topology

As large amounts of dynamic and heterogeneous resources existin gird environment, resource monitoring and discovery is one of themost important aspects in grid research. The effect of updatinginformation, the rate of finding resources successfully, the speed ofmessage transferring, the availability of resources obtained, areimportant consults to evaluate one resource management model.Resource organization and message routing affect the quality ofresource discovery. Traditional resource monitoring operated incentralized topology. Centralized topology can be implemented andmanaged easily, but has a problem that the resource database willbecome a bottleneck caused by dynamic resources attending andleaving. Its scalability and security should be also considered.New ideas are introduced to grid with the development of P2Psystem. P2P system target to share resources among users, andobviously resource management is most important in a P2P system.Naturally, collaboration of grid and P2P can be a good attempt.Recently, a lot of attempts have been tried to corporate grid with P2P.This paper analyzes current related work, illustrates advantagesand limits of them, and considers it feasible to introduce P2P to grid.Resource management is a hard research in both grid and P2P. But thecollaboration of them is not simply piecing them together. There aremany differences between them, so it is necessary to consider thisbefore integrating them.This paper, at first, introduces the concept of virtual organization(VO), and divide resources into many VOs according to their types.Thus, we present a layered resource management architecture calledC&G. There are many distributed, dynamic and heterogeneousresources in a grid, so by virtual organization, resource organizationand management become convenient. Layered architecture makesinformation searches inside local area and between organizationsrespectively independent.A super node is set inside one VO to manage all the resourcesinside the VO. The super node serves as a management center of acentralized system. Thus the performance and security can beimproved by dividing one resource center into many. Communicationsbetween organizations are through a P2P-like network formed by allthe super nodes. Then two layers are formed: one composed by alltypes of resources, and the other by super nodes. Resources queryinside virtual organizations can be satisfied by MDS of Globus.Queries in one VO are only transferred in itself just like a LAN (LocalArea Network). The C&G model achieve resource management in P2Player by P2P-like protocol. The hash algorithm of Chord is utilized inC&G to produce serial numbers from IP address. So an overlaynetwork is formed by the nodes each of which is marked by a number.This paper presents three algorithms for resource management inP2P layer: IE (Information Exchange);ID (Information Discovery);IR(Information Routing). IE updates nodes information every intervalseconds to solve problems caused by overdue resource informationand speed up resource search. IE is also the base of the other twoalgorithms. ID contributed in the situation when a node attends a grid.The new node distributes discovery messages to some other nodes toobtain information and also its information can be obtained by others.IR solves the problem how messages are routed in the network. Onegrid space can be divided into some subspaces so that messagedistribution speed can be improved by virtue of parallel search. Thethree algorithms can help a query reach more nodes. C&G makemessage complexity lower than general Gnutella-like protocol.It can be described that C&G references Chord in resourceorganization, and references Gossip in message routing. C&G hasadvantages of Chord and Gossip: Chord's organization and Gossip'sinformation query;meanwhile avoids disadvantages of them: Chord'slimitation for semantic query and Gossip's cost in bandwidth.It is feasible to introduce P2P to grid so as to organize resourcesinto a layered architecture. C&G is a good construction for resourceorganization, and the algorithms show their advantages in improvingthe quality of resource discovery. This paper can be referenced byfurther research of resource management and resource allocation, andis also a foundation for subsequent research of resource allocation.