CONNECTION CONCENTRATOR FOR DISTRIBUTED OBJECT SYSTEMS

Abstract

In a distributed object system, the services available on the network are modeled as network objects. A client typically communicates with and uses these objects using Internet connections. As the number of clients increase the servers can get overwhelmed by the number of connections coming into the server resulting in sluggish response and/or loss of service. The invention describes the mechanisms for reducing the required number of connections by automatically concentrating multiple connections onto a single connection. This is done by introducing shared intermediate connection concentrators called gateways. When the client invokes on an external object reference, the ORB running on the client automatically forwards the request to a gateway assigned to it, which then forwards it onto a shared connection to the server. The solution is symmetrical—if the server invokes on an object running inside the client, the invocation again flows through the gateway. The mechanisms described here do not require any explicit programming and can be turned ON or OFF via configuration. The disclosure also describes several policy and/or algorithm based schemes for assigning clients to gateways. Advantages of this invention include reduced interference, improved communication bandwidth, fault tolerance, modularity, scalability, and more efficient and cost-effective base stations and mobile stations.

Description

RELATED APPLICATIONS

[0001] The present application claims priority to and incorporates the following patent applications by reference:

[0002] Prov. Appl. No. 60/104,543 filed Oct. 16, 1998.

FIELD

[0003] The present invention relates to the distributed Object Oriented Systems. In particular, the invention relates to an apparatus and method for improving communication between clients and servers employing object software technology.

BACKGROUND

[0004] Conventional client-server networks employ dedicated connections between one another. In general, the client communicates directly with the server and any requests made by the client are handled by the server. This does not pose a problem in conventional networks because the number of overall connections between the clients and servers is relatively low. However, as more clients connect to more servers and as the client-server communication incorporates more object references, the number of interconnections increases dramatically. Presently, and in the future, the conventional client-server systems that incorporate object references will be overburdened and a solution to this problem must be found.

[0005] Accordingly, a limitation of the existing distributed object systems is their inability to scale to handle a large number of clients. In a distributed object system (for example, Common Object Request Broker (CORBA) (1), Java Remote Method Invocation (RMI) (2), and Distributed Component Object Model (DCOM) (3)) objects can be running all over the network. A client invokes a method on an object by first obtaining a reference to it. An object reference typically contains the network address of the server in which the object is instantiated as well as some unique datum that identifies the object within the server. When a method is invoked on the object, the client's ORB (Object Request Broker) runtime typically makes a direct connection to the server and forwards the invocation to the server (see FIG. 1) using an object protocol (for example, most CORBA implementations use Internet Inter ORB Protocol, IIOP). With this scheme, if there are N clients and M servers, there could be up to N*M connections between them (see FIG. 2). For large values of N and M, the number of connections increase beyond the capabilities of even the largest of computing systems. The result is that the object system starts to perform poorly and/or starts to reject service.

[0006] Several relevant publications are provided and incorporated herein by reference. The Common Object Request Broker (CORBA) and Internet Inter-ORB Protocol (IIOP) specifications are published by the Object Management Group at http://www.omg.com. The Java Remote Method Invocation (Java RMI) specifications are published by Sun Microsystems Java Software Division at http://wwwjavasoft.com/products/rmi. The Distributed Component Object Model is published in “Professional DCOM Programming” by Dr. R. Grimes (1997) ISBN 186100-60-X.

[0007] A goal of the invention is to overcome the identified limitations and to provide a connection concentrator for distributed object systems.

SUMMARY

[0008] A goal of the invention is to overcome the identified limitations and to provide a connection concentrator for distributed object systems. Exemplary embodiments are provided for use with the Internet, but other communication protocols can be used. This invention improves the network by incorporating gateway structures disposed between the conventional client-server architecture. The invention also provides an object request broker (ORB) at the client and the gateway so that the gateway can efficiently access objects from the servers and retrieve the object on behalf of the client. Additional clients can connect to the same gateway and access the same objects without an attendant increase in links to the server with the object.

[0009] In another embodiment of the invention, the object request broker also resides on the server and the gateway can access objects from the clients and retrieve the object on behalf of the server.

[0010] In yet another embodiment of the invention, additional gateways are disposed in series to further reduce the number of connections to the object servers.

[0011] Advantages of the invention include an increased capability of a network system to provide access to objects without overburdening the network. Additionally, the network can be further improved by placing additional gateways in series to further reduce the number of connections to the object servers.

BRIEF DESCRIPTION OF THE DRAWINGS

[0012] Additional advantages of the invention will become apparent upon reading the following detailed description and upon reference to the drawings, in which:

[0013] FIG. 1 depicts that the client ORB runtime typically opens up a direct connection to each server it needs to communicate with. The gray circles denote clients and servers, and the arrow lines denote connections;

[0014] FIG. 2 depicts that With N clients and M servers that need to communicate, there can be up to N*M number of connections;

[0015] FIG. 3 depicts that the gateway serves as an intermediate party that forwards requests from the client to the server and responses back from the server to the client. The client has only one open connection—the connection to the gateway;

[0016] FIG. 4 depicts that multiple clients are connected to a single gateway and there can be multiple gateways. If there are L gateways, the maximum number of connections is now N+L*M. Since L is much smaller than N, the connection resources used with this scheme is dramatically reduced; and

[0017] FIG. 5 depicts that the gateways can be stacked to further increase the number of clients that can be simultaneously connected to the servers.

DETAILED DESCRIPTION

[0018] The exemplary embodiments are described herein with reference to specific configurations and protocols. Those skilled in the art will appreciate that various changes and modifications can be made to the exemplary embodiments while remaining within the scope of the present invention.

[0019] Referring to FIG. 3, the invention provides an improves network system by using shared intermediate connection concentrators called gateways to deliver the invocation from the clients to the servers and responses back from the servers to the clients. A gateway channels requests and responses from multiple clients to and from the server respectively over a single connection to the server (see FIG. 3). The use of intermediate gateways is done in such a way as to not require any change to the client or server programs.

[0020] When the client initializes its ORB runtime, the ORB contacts a gateway factory on the network and obtains a gateway object. When an external object reference is received (or read) by a client ORB runtime, it creates a wrapped (or a proxy) object references such that it now contains the network address of the gateway and the datum contains the original datum as well as the original address of the server. When the client invokes a method on this object reference, the invocation now automatically goes to the gateway instead of the server. The gateway unwraps the datum to get the original datum and the original address of the server. It then delivers the request to the server. The gateway is also responsible for returning the response from the server back to the client. A gateway is shared by multiple clients and there can be multiple gateways that are managed by a gateway factory, as shown in FIG. 4.

[0021] A gateway may have to perform some protocol specific transformation on the messages that pass through it. For example, if the protocol is Internet Inter ORB Protocol (IIOP) version 1.1, then the gateway needs to manipulate the IIOP request, response, locate request, and locate response headers as follows.

[0022] For IIOP request and IIOP locate request, the request id in the messages from the clients needs to be manipulated to make them unique on the connection to the server. This transformation has to be undone when the IIOP response or IIOP locate response is received by the gateway from the server before the response is returned to the client.

[0023] 1) The IIOP request and locate request contain the wrapped datum (the object key). This needs to be unwrapped and set to the original datum.

[0024] 2) The above transformations may change the byte alignment of the request message payload. The gateway needs to make sure that the message payload is properly byte aligned. Padding its header with the required number of bytes is one possible solution.

[0025] The above solution is symmetrical. If a client is also a server, then, when it sends (or writes) references for object instantiated in itself to some one else, the object reference is again wrapped with the address of the gateway. As a result, when a method is invoked on the object, the invocation again automatically flows through the gateway. That is, all traffic to and from a program is channeled over a single connection (or two connections, if an asymmetrical protocol like IIOP is used where requests only flow in one directions). This results in a far lower utilization of the connection resources, and hence the object system can scale to support a far greater number of clients and servers.

[0026] In another embodiment, as shown in FIG. 5, the gateways can be connected in series (stacked) to further increase the number of clients that can be simultaneously connected to the servers.

[0027] The gateway factory can use different algorithms or policies to assign gateway objects to the clients, for example:

[0028] 1) round-robin assignment—the gateway factory round robins through the list of gateway objects that it manages, distributing the number of clients evenly to them;

[0029] 2) locality based assignment—the gateway factory assigns a gateway that is closest to the client; and/or

[0030] 3) load-based assignment—the gateway factory assigns the least loaded gateway, where load can be defined as some function of the number of clients and the number of requests flowing through the gateway.

[0031] Since clients may disconnect after some time, the gateway factory may periodically query the gateways to determine the number of clients connected to it. This helps the gateway factory in its future assignment decisions.

[0032] The above mechanisms can also be used in conjunction with public key based connection level authentication protocols like Secure Sockets Layer (SSL.) To transmit the identity of the client to the server, the server and the gateway are in a trust relationship such that

[0033] 1. The client authenticates itself to the gateway;

[0034] 2. The gateway authenticates itself to the server;

[0035] 3. The gateway passes along the identity of the client to the server on each request; and

[0036] 4. The server, relying on the trust relationship with the gateway, assumes the real identity of the caller to be the one passed with the request by the gateway.

[0037] Advantages of the present invention include reduced interference, improved communication bandwidth, fault tolerance, modularity, scalability, and more efficient and cost-effective base stations and mobile stations.

[0038] Having disclosed exemplary embodiments and the best mode, modifications and variations may be made to the disclosed embodiments while remaining within the scope of the present invention as defined by the following claims.

Claims

1. An apparatus for supporting an object based distributed software system, comprising:

a plurality of clients each running an object request broker program;

a plurality of gateways each running an object request broker program, coupled to the clients and configured to communicate with the clients, and in at least one instance to receive a proxified object request from a client and to decode the proxified object request to generate an object request; and

a plurality of servers coupled to the gateways and configured to communicate with the gateways, and in at least one instance to receive an object request and to provide the requested object to the gateway making the request.

2. The apparatus of

claim 1, wherein:

each of the plurality of servers also run an object request broker program.

3. The apparatus of

claim 1, whereby:

the apparatus tremendously increases the number of clients that can be simultaneously connected to an object server or a group/cluster of object servers

4. The apparatus of

claim 1, whereby:

programs can be written without any knowledge of the mechanisms described here and still enjoy all the benefits just through changed configuration settings.

5. The apparatus of

claim 1, wherein:

the scheme is symmetrical, and when an object server acting as a client communicates back to an object running in the client, it still uses the gateway to which the client is then currently connected

6. The apparatus of

claim 1, wherein:

the scheme provides support for doing all sorts of static and dynamic assignment of clients to gateways based on policies and/or algorithms

7. The apparatus of

claim 1, wherein:

the scheme provides support for client authentication over public key based connection level authentication mechanisms