Scalable resource discovery and reconfiguration for distributed computer networks

Abstract

A method is provided for discovering resources in a network of user nodes. According to the method, a resource request to be published is received at a first user node of the network, and it is determined (e.g., randomly) whether or not to send the resource request to a server node. When it is determined not to send the resource request to the server node, the resource request is forwarded to a second user node of the network through a direct connection. When it is determined to send the resource request to the server node, the resource request is sent to the server node for publication. Also provided is a user node for use in a computer network of the type that includes user nodes and at least one server node, with each user node being connected to at least one other user node through a direct connection.

Description

CROSS-REFERENCE TO RELATED APPLICATIONS

[0001] This application is related to the inventor's application “SYSTEM AND METHOD FOR RESPONDING TO RESOURCE REQUESTS IN DISTRIBUTED COMPUTER NETWORKS,” Ser. No. ______, now ______, which was filed on the same day as the present application and commonly assigned herewith to International Business Machines Corporation. This related application is incorporated herein by reference.

BACKGROUND OF THE INVENTION

[0002] 1. Field of the Invention

[0003] The present invention relates to computer networks, and more specifically to a framework for scalable resource discovery and dynamic reconfiguration in distributed computer networks.

[0004] 2. Description of Related Art

[0005] Computer networks such as the Internet allow users to share resources such as files and hardware. The expansion of the Internet and the adoption of standards for the World Wide Web have made the viewing and downloading of files by a user almost effortless. The user need not know any programming languages. By simply running an Internet browser, the user only needs to point and click to view and download desired files. The availability of such programs allows for easy collaboration and file sharing among like-minded individuals separated by great distances over a distributed computer network, which can literally span the entire globe.

[0006] Conventionally, a distributed computer network is set up to have a client/server framework. In particular, each user is a client that can access a server node over the network and, with the proper authorization, publish files to the server node. Once a file is published to the server node, other clients on the network can access the server node to view or download the file. Additionally, the server node can allow a client to automatically send a file to another client that is reachable over the network. The client simply sends the file to the server node along with information identifying the desired recipient, and the server node sends the file on to the corresponding client. The server node can also be used to allow the clients to share hardware resources such as a printer.

[0007] With such a client/server framework, the server node is charged with providing security. For example, the server node must insure that only authorized clients can use the network resources (e.g., download files), and that only proper files are published. Additionally, the server node represents a single point of failure. Thus, in any client/server environment in which reliability is required, the server node must be of industrial strength and have redundant systems to prevent system shutdowns and data loss. Further, because all client-to-client resource transfers pass through the server node, the adding of another client to the network puts an additional burden on the server node and degrades network performance.

[0008] In such a client/server framework, the clients have little privacy. Typically, the server node requires authentication before allowing a client to access network resources. Once the client has provided authentication credentials, the server node can easily log all of the network activity of the client. For example, the server node could keep a log of all files uploaded and downloaded by the client. Even if access by unauthenticated clients is allowed, the server node can use any of various unique identification techniques to track client activity over time. For example, the server node can place a unique cookie on the client and later use the cookie to identify the client each time it accesses the server node.

[0009] One solution to some of the drawbacks of the conventional client/server framework is provided by a “viral” network. In such a network, a user node connects to one or more known hosts that are participating in a highly interconnected virtual network. Then, the user node itself becomes a host node that can respond to requests for resources and available hosts. Each user in the network forwards resource requests to all known neighboring nodes, so as to potentially propagate each request throughout the entire network. For example, the Gnutella system employs such a viral network framework. Gnutella has a published network protocol and provides users a client/server application (available at http://gnutella.wego.com) that allows each user to act as a host node in a file sharing network. The Gnutella system can be used to securely distribute commercial content that is protected by encryption and licensing.

[0010] Viral networks are based on peer-to-peer communication. Peer-to-peer is a communications model in which each party has similar capabilities and either party may initiate a communication session. For example, the Gnutella application employs peer-to-peer communication to allow users to exchange files with one another over the Internet. The peer-to-peer model used in a viral network relies on each peer (i.e., user node) having knowledge of at least one of the other peers in the network. When searching for a resource such as a file, a peer sends a resource request to other known peers, which in turn pass it on to their known peers and so on to propagate the request throughout the network. A peer that has the resource and receives the request can send the resource (or a message indicating its availability) back to the requesting peer. Because such a framework offers independence from a centralized network authority (e.g., server node), users in a viral network have enhanced privacy and the single point of failure is eliminated.

[0011] FIG. 1 shows an exemplary viral network. Each node in the network represents a user that acts both as a client and host, and is connected with one or more other nodes. When a first node 210 desires a particular resource (e.g., file), the first node 210 issues a request to all known nodes 202, 204, 206, and 208, which in turn do the same. For example, the request reaches a second node 212 by being passed in succession through nodes 208, 216, and 218. If the second node 212 has the requested resource, it responds by sending an appropriate message to the first node 210 (e.g., back the same path that the request traversed). Because a node having the requested resource has been identified, the first node 210 can initiate a direct peer-to-peer connection with the second node 212 in order to download the resource. Throughout the viral network, any number of such resource requests, acknowledgments, and transfers can occur simultaneously.

[0012] While viral networks offer enhanced privacy and eliminate a single point of failure, the framework has drawbacks related to scalability. In a large, decentralized viral network, efficient resource discovery breaks down as the number of participating nodes increases. More specifically, are source request can only propagate from node to node, and each node only propagates the request to a relatively small number of other nodes. To control network traffic and prevent unreasonable response times, a practical system must employ a “time-to-live” or some limit on the number of times a request can be forwarded (i.e., a maximum number of peer hops). This effectively disconnects any two nodes or groups of nodes that are separated by a path that would require a request to propagate through an unreasonably large number of intermediary nodes. Further, any such limit on request propagation makes it impossible to perform an exhaustive search for a resource, because such a search would require the request to be propagated to all of the nodes in the network.

[0013] Additionally, there has recently been proposed a content-based publish-subscribe messaging infrastructure that utilizes an information flow graph. For example, the Gryphon system (described at http://www.research.ibm.com/gryphon) has been developed by the assignee of the present invention. This system provides a content-based subscription service and performs message brokering by merging the features of distributed publish/subscribe communications and database technology. At the core of the Gryphon system is an information flow graph that specifies the selective delivery of events, the transformation of events, and the generation of new events.

[0014] FIG. 2 shows an exemplary content-based publish-subscribe messaging infrastructure that utilizes information flow graphs. In this system, stocks trades derived from two information sources NYSE and NASDAQ are combined, transformed, filtered and delivered to subscribing clients. For example, one user 312 may subscribe to the message-brokering server 302 and request to receive all stock trades on both the NYSE and NASDAQ that have a value of over one million dollars. The message broker 302 receives raw stock trade information such as price and volume from the NYSE 324 and NASDAQ 326.

[0015] Based on the information request of the user 312, the server 302 merges the stock trade information from the two sources, transforms the raw price and volume information into value information for each trade, and then filters the derived values to produce the subset of trades that are valued at over one million dollars. In a similar manner, each subscribing user (e.g., nodes 304, 306, and 308) specifies its own criteria, and the message-brokering server 302 performs information selection, transformation, filtering, and delivery in order to provide each user with the requested information.

[0016] While the publish-subscribe messaging infrastructure of FIG. 2 provides good scalability for a messaging system with a large number of users, as in the conventional client/server framework the users have little privacy. All users must identify themselves when subscribing to the system and all information is delivered to the user through the centralized server. Thus, the centralized server can easily maintain a log of all users of the system and the exact information that each desires and receives. The centralized message-brokering server also represents a single point of failure for the system.

SUMMARY OF THE INVENTION

[0017] In view of these drawbacks, it is an object of the present invention to remove the above-mentioned drawbacks and to provide a network framework that provides scalable resource discovery and sharing. A scalable messaging infrastructure is provided to address scalability and performance issues, while most features of a decentralized network are preserved. At least one publish-subscribe server node is provided in a decentralized network not as a central authority, but as a messaging infrastructure layer. Thus, scalability can be achieved in a decentralized network.

[0018] One embodiment of the present invention provides a method for discovering resources in a network of user nodes. According to the method, a resource request to be published is received at a first user node of the network, and it is determined whether or not to send the resource request to a server node. When it is determined not to send the resource request to the server node, the resource request is forwarded to a second user node of the network through a direct connection. When it is determined to send the resource request to the server node, the resource request is sent to the server node for publication. In a preferred embodiment, the determination of whether or not to send the resource request to the server node is a random decision made by the first user node.

[0019] Another embodiment of the present invention provides a user node for use in a computer network of the type that includes user nodes and at least one server node, with each of the user nodes being connected to at least one other user node through a direct connection. The user node includes a receiving interface for receiving a resource request to be published, control means for deciding whether or not to send the resource request to the server node, and at least one transmitting interface for selectively forwarding the resource request to a second user node of the network through a direct connection or sending the resource request to the server node for publication. The transmitting interface forwards the resource request to the second user node when the control means decides not to send the resource request to the server node, and sends the resource request to the server node for publication when the control means decides to send the resource request to the server node. In one preferred embodiment, the control means randomly selects one of the other user nodes of the network to be the second user node to which the resource request is forwarded.

[0020] Other objects, features, and advantages of the present invention will become apparent from the following detailed description. It should be understood, however, that the detailed description and specific examples, while indicating preferred embodiments of the present invention, are given by way of illustration only and various modifications may naturally be performed without deviating from the present invention.

BRIEF DESCRIPTION OF THE DRAWINGS

[0021] FIG. 1 is a diagram of an exemplary viral network;

[0022] FIG. 2 is a diagram of an exemplary content-based publish-subscribe messaging infrastructure;

[0023] FIG. 3 is a diagram of scalable network framework according to a preferred embodiment of the present invention;

[0024] FIG. 4 is a flow chart of a process for obtaining a resource within a scalable network framework in accordance with a first embodiment of the present invention;

[0025] FIG. 5 is a flow chart of a process for obtaining a resource within a scalable network framework in accordance with a second embodiment of the present invention;

[0026] FIG. 6 is a flow chart of a process for obtaining a resource within a scalable network framework in accordance with a third embodiment of the present invention;

[0027] FIG. 7 is a diagram of scalable network framework showing an exemplary implementation of part of the process of FIG. 6; and

[0028] FIG. 8 is a diagram of scalable network framework showing an exemplary implementation of another part of the process of FIG. 6.

DETAILED DESCRIPTION OF PREFERRED EMBODIMENTS

[0029] Preferred embodiments of the present invention will be described in detail hereinbelow with reference to the attached drawings.

[0030] FIG. 3 shows a scalable network framework according to a preferred embodiment of the present invention. As shown, the framework includes a publish-subscribe server node 402 and multiple user nodes (e.g., 404, 406, and 410). In embodiments of the present invention, the server node 402 can be implemented by a single server or by a “server cloud” that is made up of any number of servers. The individual servers of such a server cloud can be connected to one another and to the Internet in various ways and can even be separated by great distances so as to provide an appropriate level of service and advantageous features such as data and path redundancy.

[0031] Within this framework, a user node 416 joins the network by contacting the publish-subscribe server node 402 and subscribing to certain “channels” of messages. Additionally, the joining user node 416 solicits connections and then directly connects with at least one other user node 410, 418, and 420 (e.g., based on some criteria such as geographic location, connection speed, or common interests). In this manner, all of the user nodes in the network are connected to the centralized publish-subscribe server node for messaging (not shown for clarity) and to one another through peer-to-peer connections that form a decentralized viral-type resource sharing network.

[0032] FIG. 4 is a flow chart of a process for obtaining a resource within such a scalable network framework in accordance with a first embodiment of the present invention. Whenever a first user node 416 in the network desires are source (e.g., file), are source request (or query) is sent to the server node 402 (step S10), and the server node 402 publishes the resource request by sending it to all of the user nodes that are subscribed to the channel corresponding to that type of request (step S12). A second user node 422 that receives the request and is willing to provide the resource contacts the first user node 416 (step S14), and the first and second user nodes 416 and 422 set up a peer-to-peer connection to provide the requested resource to the first user node 416 (step S16). In this manner, the publish-subscribe messaging infrastructure layer allows a resource request to reach nodes that are separated from the requesting node by a direct-connect path that includes a very large number of intermediary nodes.

[0033] Thus, in the scalable network framework provided by the present invention, efficient resource discovery is maintained when the number of user nodes in the network increases. At the same time, the resources themselves are not published to the server and the actual resource sharing does not involve the server, so the demands on the server are less than in the conventional client/server framework. Furthermore, because each search request is published to all of the user nodes that subscribe to the relevant channel, it is possible to perform an exhaustive (or at least very extensive) search for the requested resource within an acceptable time frame in a network that contains a very large number of user nodes.

[0034] In preferred embodiments, the publish-subscribe messaging infrastructure supports two types of “channels”, or shared data streams. The first type of channel is a “node discovery channel” that is used to support the discovery of other user nodes in the decentralized network. To join the network, a user node announces itself using the publish-subscribe messaging infrastructure of the server node. In particular, a join announcement from the new user node is routed via the server node to one or more of the node discovery channels to solicit connections from other user nodes. Thus, the publish-subscribe infastructure functions as a general purpose transport layer (like a layer above IP) that establishes connections to other user nodes to accomplish the broadcasting of join announcements.

[0035] Such join announcements can be divided among individual node discovery channels based on categories such as geographic location, network connection speed, types of resources available, and/or common interests. Thus, the new user node can discover remote user nodes (both geographically and with respect to network hops) and when connecting can favor nodes with the desired type of resources or acceptable connectivity attributes. Further, because each join announcement is only sent to the user nodes subscribed to certain channels, the user nodes of a very large network are not constantly flooded with join announcements.

[0036] The second type of channel is a “resource request channel” that is used to optimize the publication of resource requests. As explained above, the server node publishes each resource request by sending it to all of the user nodes that are subscribed to a particular channel (or channels). These resource request channels can be divided into a rich taxonomy of resource types in order to allow the user nodes to effectively filter out undesired requests. Further, rather than reaching every user node in the network, the use of resource request channels allows a resource request to only reach a relevant subset of user nodes (i.e., those that subscribe to the relevant channel or channels).

[0037] Thus, a user node that only desires to share technical documentation is not bombarded with requests for multimedia content. Depending on the application, the user node and/or the server node can determine the channel or channels on which to publish an individual announcement or resource request. Further, in preferred embodiments, a channel exists merely by virtue of being published to by a user node. Thus, any user node can create anew resource channel and then promote its availability for subscription in any conventional manner (e.g., through a web site, news group, television, or direct mail advertising).

[0038] FIG. 5 is a flow chart of a process for obtaining a resource within a scalable network framework in accordance with a second embodiment of the present invention. Whenever a first user node 416 in the network desires a resource (e.g., file), a resource request is sent to the server node 402 (step S10), and the server node 402 publishes the resource request by sending it to all of the user nodes that are subscribed to the channel corresponding to that type of request (step S12). Additionally, in the second embodiment the first node 416 also sends the resource request to all of the user nodes 410, 418, and 420 to which it is connected in the decentralized network.

[0039] This process is repeated with each user nodes that receives the request passing it on to the user nodes to which it is connected, so as to propagate the request through the user nodes of the decentralized network (step S11). For example, the request reaches a second user node 404 by being passed in succession through nodes 410 and 414. Further, in some embodiments each user node that receives the resource request from the server node (e.g., user node 422) also sends the resource request to all of the user nodes to which it is connected in the decentralized network. If the second user node 404 receives the request (either through user node propagation or from the server node) and is willing to provide the resource, the second user node 404 contacts the first user node 416 (step S14), and a peer-to-peer connection is set up to share the requested resource (step S16).

[0040] Thus, in the second embodiment resource requests are both published through the messaging infrastructure layer (as in the first embodiment) and propagated through the user nodes of the decentralized network. Because this dual path process offers an alternative to the request propagation path that passes through the centralized server node, the single point of failure is eliminated. In other words, the resource request is propagated to other user nodes even when the server node is down. Further, in the second embodiment, it is not necessary for all of the user nodes to be connected to, or even know about the presence of the publish-subscribe infrastructure. Such a user node can merely forward resource requests to all of the user nodes to which it is connected. This feature allows the second embodiment to be implemented in an existing network without requiring the modification of all user nodes.

[0041] FIG. 6 is a flow chart of a process for obtaining a resource within such a scalable network framework in accordance with a third embodiment of the present invention. In this embodiment, resource requests are not sent directly to the server node in order to give enhanced privacy to the requesting user node. Whenever a resource (e.g., file) is desired in the third embodiment, the requesting first user node 410 sends a resource request to a second user node 414 to which it is connected in the decentralized network (step S20), as shown by the dashed arrows in FIG. 7. The second user node 414 then determines whether to send the request to the publish-subscribe server node 402 or to another user node to which it is connected in the decentralized network (step S22). Preferably, all of the user nodes are connected to the centralized publish-subscribe server node for messaging (not shown for clarity).

[0042] If it is decided not to send to the server node 402, then the second user node 414 forwards the resource request to another user node 420 to which it is connected (step S20), as shown in FIG. 7. This forwarding process is repeated with each user node that receives the resource request making the same determination (step S22). When a third user node 416 decides to send to the server node 402, the resource request is sent to the server node 402 (step S24), and the server node 402 publishes the resource request by sending it to all of the user nodes that are subscribed to the channel corresponding to that type of request (step S26).

[0043] In preferred embodiments, each determination of whether or not to send the request to the publish subscribe server node is a “random” decision made by the user node based on a weighting factor of between 0 and 1 that gives the probability that the request will be sent to the server node. For example, if the weighting factor is 0.25, then there is a 25% chance that the user node will send the request to the server node and a 75% chance that the request will be forwarded to another user node. Thus, on average a weighting factor of 0.25 should cause resource requests to be forwarded to other user nodes three times before being sent to the publish-subscribe server node. The value of the weighting factor is set based on factors such as the desired level of privacy. Further, the weighting factor can be a fixed value that is used throughout the network or can be set by the user nodes on a per message basis. Other criteria such as a maximum number of forwards or a maximum elapsed time can also be incorporated into the determination that is made by each user node receiving the request.

[0044] In further embodiments, the determination of whether or not to send the request to the publish-subscribe server node is made based on some other criteria such as a fixed number of forwards. For example, in one embodiment each resource request is always forwarded through three user nodes and then sent to the publish-subscribe server node. Preferably, whenever a resource request is to be forwarded on to another user node, the choice of which other user node will receive the forwarded request is made through a random selection of one of the user nodes to which the forwarding user node is connected.

[0045] After the resource request is published by the server node 402, a fourth user node 422 that receives the request and is willing to provide the resource contacts the first user node 410. More specifically, as shown in FIG. 8 the fourth user node 422 sends a response to the third user node 416 (step S28), which had sent the resource request to the server node 402 for publication. The third user node 416 forwards the response to the user node 420 from which it received the resource request, and this forwarding process is repeated until the response reaches the first user node 410 (step S30). At this point, the first user node 416 can contact the fourth user node 422 to set up a direct peer-to-peer connection to share the requested resource (step S32).

[0046] In preferred embodiments, the response from the user node having the resource contains metadata concerning the resource match. When the response is received, the requesting user node evaluates the metadata and then decides whether to make a direct connection with the responding node to share the resource itself(e.g., download the requested file). If multiple responses are received, the requesting user node can evaluate the metadata in each response and then select one or more of the responding user nodes based on any criteria (e.g., past experience, order received, connection speed, or physical location).

[0047] Additionally, in preferred embodiments, responses from user nodes having the resource are routed through existing connections (such as the one the request had traversed through the server node), and a new point-to-point connection is only established to actually transfer the resource. Otherwise, with each responding user node initiating a new point-to-point connection, the user node receiving the responses could be overwhelmed whenever a large number of responses are received. Alternatively, the server node could set up a matching “one-time” response channel or a permanent response channel to be used by user nodes responding to the resource request. The publishing of responses over a permanent response channel would facilitate “passive” user nodes that archive responses of possible interest for future use.

[0048] Because resource requests are forwarded through one or more other user nodes rather than being sent directly to the server node, the third embodiment offers privacy to requesting user nodes. The actual user node requesting a resource remains anonymous to the server node, so the server node cannot keep track of which users are sharing (or even requesting) which resources. As in a conventional viral network, in the third embodiment only a user node that actually provides the resource has knowledge of the resource sharing and the identity of the requesting node. Further, unlike a conventional viral network, in the third embodiment the use of a publish-subscribe messaging infrastructure layer allows for efficient resource discovery in a network having a very large number of user nodes. Thus, the present invention allows scalability to be achieved in a decentralized network while enhanced user privacy is maintained.

[0049] The embodiments of the present invention described above require some mechanism for identifying individual messages. In preferred embodiments, every message (i.e., resource request or response) is assigned a unique identification number. For example, one embodiment employs an algorithm developed by Microsoft Corporation that allows each user node to individually generate globally-unique message identifiers (GUIDs) that are very likely to be globally unique. Additionally, each user node must store (e.g., in a table) at least a limited history of forwarded messages in order to allow responses (possibly including the resource itself) to reach the user node that sent a message through the same path. Further, some embodiments include a mechanism to prevent the looping of a resource request. For example, the globally-unique message identifiers GUIDs and node history tables can easily be used to create an anti-looping mechanism.

[0050] While the embodiments of the present invention described above relate to a single server node, multiple publish-subscribe server nodes can be provided in the network to further minimize “point of failure” concerns. Further, competing providers may coexist in the network by operating different server nodes and competing for user node subscriptions. Additionally, the features of the different embodiments described above can be combined for further applications. For example, one embodiment of the present invention includes both the randomized request forwarding of the third embodiment and the dual propagation/publication of the second embodiment. Other design choices, such as network protocols, forwarding criteria, and membership criteria, could easily be adapted.

[0051] The present invention can be realized in hardware, software, or a combination of hardware and software. Any kind of computer system—or other apparatus adapted for carrying out the methods described herein—is suited. A typical combination of hardware and software could be a general purpose computer system with a computer program that, when loaded and executed, controls the computer system such that it carries out the methods described herein.

[0052] The present invention can also be embedded in a computer program product, which comprises all the features enabling the implementation of the methods described herein, and which—when loaded in a computer system—is able to carry out these methods. In the present context, a “computer program” includes any expression, in any language, code or notation, of a set of instructions intended to cause a system having an information processing capability to perform a particular function either directly or after either or both of the following: a) conversion to another language, code, or notation; and b) reproduction in a different material form.

[0053] Each computer system may include one or more computers and a computer readable medium that allows the computer to read data, instructions, messages, or message packets, and other computer readable information from the computer readable medium. The computer readable medium may include non-volatile memory such as ROM, Flash memory, a hard or floppy disk, a CD-ROM, or other permanent storage. Additionally, a computer readable medium may include volatile storage such as RAM, buffers, cache memory, and network circuits. Furthermore, the computer readable medium may include computer readable information in a transitory state medium such as a network link and/or a network interface (including a wired network or a wireless network) that allow a computer to read such computer readable information.

[0054] While there has been illustrated and described what are presently considered to be the preferred embodiments of the present invention, it will be understood by those skilled in the art that various other modifications maybe made, and equivalents maybe substituted, without departing from the true scope of the present invention. Additionally, many modifications maybe made to adapt a particular situation to the teachings of the present invention without departing from the central inventive concept described herein. Furthermore, an embodiment of the present invention may not include all of the features described above. Therefore, it is intended that the present invention not be limited to the particular embodiments disclosed, but that the invention include all embodiments falling within the scope of the appended claims.

Claims

1. A method for discovering resources in a network of user nodes, said method comprising the steps of:

receiving a resource request to be published at a first user node of the network;

determining whether or not to send the resource request to a server node;

forwarding the resource request to a second user node of the network through a direct connection, when it is determined not to send the resource request to the server node; and

sending the resource request to the server node for publication, when it is determined to send the resource request to the server node.

2. The method as defined in claim 1, wherein in the determining step, the determination of whether or not to send the resource request to the server node is a random decision made by the first user node.

3. The method as defined in claim 2, wherein in the determining step, the random decision is made based on a weighting factor corresponding to the probability that the first user node will decide to send the resource request to the server node.

4. The method as defined in claim 1, wherein the forwarding step includes the sub-steps of:

randomly selecting one of the user nodes to which the first user node is connected to be the second user node; and

forwarding the resource request from the first user node to the second user node through a direct connection.

5. The method as defined in claim 1, further comprising the step of publishing the resource request to at least some of the user nodes of the network via the server node.

6. The method as defined in claim 5, wherein in the publishing step, the server node publishes the resource request to all of the user nodes of the network that are subscribed to one or more selected resource request channels.

7. The method as defined in claim 1, further comprising the step of repeating the steps of determining and forwarding until in the determining step a user node that received the resource request decides to send the resource request to the server node.

8. The method as defined in claim 1, further comprising the steps of:

sending the resource request to be published from a requesting user node, which desires the request resource, to the first user node; and

sending an identical resource request from the requesting user node to all of the user nodes to which the requesting user node is connected through direct connections.

9. A machine-readable medium encoded with a program for discovering resources in a network of user nodes, said program containing instructions for performing the steps of:

receiving a resource request to be published at a first user node of the network;

determining whether or not to send the resource request to a server node;

forwarding the resource request to a second user node of the network through a direct connection, when it is determined not to send the resource request to the server node; and

sending the resource request to the server node for publication, when it is determined to send the resource request to the server node.

10. The machine-readable medium as defined in claim 9, wherein in the determining step, the determination of whether or not to send the resource request to the server node is a random decision made by the first user node.

11. The machine-readable medium as defined in claim 10, wherein in the determining step, the random decision is made based on a weighting factor corresponding to the probability that the first user node will decide to send the resource request to the server node.

12. The machine-readable medium as defined in claim 9, wherein the forwarding step includes the sub-steps of:

randomly selecting one of the user nodes to which the first user node is connected to be the second user node; and

forwarding the resource request from the first user node to the second user node through a direct connection.

13. The machine-readable medium as defined in claim 9, wherein said program further contains instructions for performing the step of publishing the resource request to at least some of the user nodes of the network via the server node.

14. The machine-readable medium as defined in claim 13, wherein in the publishing step, the server node publishes the resource request to all of the user nodes of the network that are subscribed to one or more selected resource request channels.

15. The machine-readable medium as defined in claim 9, wherein said program further contains instructions for performing the step of repeating the steps of determining and forwarding until in the determining step a user node that received the resource request decides to send the resource request to the server node.

16. The machine-readable medium as defined in claim 9, wherein said program further contains instructions for performing the steps of:

sending the resource request to be published from a requesting user node, which desires the request resource, to the first user node; and

sending an identical resource request from the requesting user node to all of the user nodes to which the requesting user node is connected through direct connections.

17. A user node for use in a computer network of the type that includes a plurality of user nodes and at least one server node, with each of the user nodes being connected to at least one other user node through a direct connection, said user node comprising:

a receiving interface for receiving a resource request to be published;

control means for deciding whether or not to send the resource request to the server node; and

at least one transmitting interface for selectively forwarding the resource request to a second user node of the network through a direct connection or sending the resource request to the server node for publication,

wherein the transmitting interface forwards the resource request to the second user node when the control means decides not to send the resource request to the server node, and sends the resource request to the server node for publication when the control means decides to send the resource request to the server node.

18. The user node as defined in claim 17, wherein the control means randomly decides whether or not to send the resource request to the server node.

19. The user node as defined in claim 18, wherein the control means randomly decides based on a weighting factor.

20. The user node as defined in claim 17, wherein the control means randomly selects one of the other user nodes of the network to be the second user node to which the resource request is forwarded.