US 6,145,013 AGrant
Distributed Instantiation System and Method
Issue Date:2000-11-07
•8 Claims
•16 Drawing Sheets
Abstract
A communication system and method includes unums distributed over at least a single presence and including a selected plurality of ingredients. An unum is established by creating ingredients at the level of its interface and attributes; and at the level of its implementation; and interconnecting ingredients into presences and unums. Communication between ingredients is accomplished within a single presence, across an unum boundary within an agency, or within a single unum across a presence boundary. Trust boundaries are established between presences and unums to establish a predetermined level of communications security in messaging between ingredients.
Metadata
Assignee
- Communities.com
Inventors
- Chip Morningstar
- F. Randall Farmer
Application Information
Application Number:US 811756&
Filing Date:1997-03-06
Priority Date:1996-04-09
Art Unit:275
Classifications
IPC:
G06F 946
Field of Search:
709300;303
Patent Drawings (16 sheets)
Description
Copyright Notice
A portion of the disclosure of this patent document contains material which is subject to copyright protection. The copyright owner has no objection to the xerographic reproduction by anyone of the patent document or the patent disclosure in exactly the form it appears in the Patent and Trademark Office patent file or records, but otherwise reserves all copyright rights whatsoever.
Background of the Invention
The present invention relates generally to systems and methods for communicating information in a distributed information system and, in a specific embodiment, provides an improved system for handling virtual objects in a distributed system.
The term "object" has taken on multiple independent meanings through overuse. In object-oriented programming, an object is a programming "package" of data elements and executable code. In a virtual world environment, "object" refers to a construct which is a definite entity, usually modelling an analogous real-world object (a third meaning of the term "object"). To further confuse the terminology, virtual world objects modelling real-world objects are often implemented using OOP objects. Therefore, less confusing terminology is used herein. "Object" refers to a model of a real-world object, such as a rock. "Server" refers to an OOP object.
Several other terms used herein have understood meanings in the field of object-oriented programming. A server encapsulates data and methods (executable code for performing functions specific to that server). The template for that server is called its class. For example, a server for a directory listing might encapsulate (as the data) the directory entries and (as the methods) routines for printing the list, sorting the list, etc. specific to that class. A class does not actually contain any data. The data for an actual server exists only when an "instance" of the server is instantiated. The code for the methods and the structure of the data of a server is described by the class definition for the server's class.
Often, the need exists for an instance of an object to be available in different places, on different machines running different operating systems and different hardware, and for a model of the environment which allows for an object to be distributed over multiple machines.
Currently known distributed objects are organized in a hierarchy controlled at a server node. Such a hierarchical distributed object organization is inflexible, because all control is exercised by a server. For network participants not controlling the server, the result is operational impotence.
Currently, individuals developing object oriented distributed systems may follow a model in which state information with respect to selected objects of a distributed system is replicated in each object with respect to selected other objects of a particular distributed system. Another approach to developing object-oriented distributed systems follows a model in which state information with respect to selected objects of a distributed system must be explicitly communicated for each object with respect to selected other objects of a particular distributed system. Either approach is accompanied by undesired overhead which hampers operability. The first approach has undue overhead due to excessive replication of state information. The second approach is hampered by excessive communications overhead involving objects obtaining state information from each other. It is accordingly desirable to develop a distributed object-oriented system which optimizes the sharing and retention of state information between objects which are established for interoperability with each other.
Summary of the Invention
According to an embodiment of the present invention, an object-oriented system includes distributed objects having presences over a plurality of associated agencies. The agencies have agency boundaries such that one agency need not trust another agency to operate correctly. Unums represent entities in the object-oriented distributed system, where an unum has a presence on each of multiple agencies where other objects on those agencies desire to interact with the unum.
Each presence of the unum is organized into logical units called "ingredients". An ingredient includes state information and code implementing the particular presence associated with the unum distributed over several agencies. The state information linking the related object ingredients includes replicated references associating object ingredients as belonging to presences on particular agencies. The ingredients belonging to a common unum communicate directly with each other over agency boundaries by virtue of a trust boundary traversing agency boundaries. Similarly, ingredients belonging to a common agency communicate directly with each other over unum boundaries by virtue of a trust boundary traversing unum boundaries. The various trust boundaries are established by reference state information in each ingredient object. The references establish the allegiance of a particular ingredient to a particular unum and to a particular agency. The references thus make clear the allegiance of a particular ingredient to a particular unum.
The advantage of organizing ingredients according to unums distributed over several agencies is that an operable distributed system of users are able to communicate with object-oriented messages without having to engage in explicit machine-to-machine or agency-to-agency communication. By becoming a member of an unum system, simple, direct intercommunication is enabled without having to communicate explicitly across machine or agency boundaries.
An ingredient interface according to the present invention defines a message protocol that the ingredient presents to other ingredients within the same presence. The ingredient state descriptor defines state information internal to the ingredient which is instantiated for each instance of the presence which contains the ingredient. An ingredient implementation is the realization of the ingredient on a selected platform (e.g., Windows) or in terms of some particular standard (e.g., JPEG). Ingredient state information according to the present invention is maintained on a per-instance basis. The state of a particular ingredient is contained in a set of typed, named fields, each of which functions as a state variable. All the ingredients of all the presences of a particular unum collectively maintain the total state of the unum. The state maintained in a presence includes a joint state and a local state. An ingredient contributes to one of these two kinds of state information. Such ingredients are called joint ingredients and local ingredients, respectively. The local state according to the present invention is known only to the particular agency on which that presence is found and is not shared with others. A field of a local ingredient is visible only to that ingredient. The joint state according to the present invention is known to all presences of the unum and is shared in common among them. While the fields of a joint ingredient are scoped in the same way the fields of a local ingredient are, a copy of any given joint ingredient is found in all presences, and its state is the same in all presences (neglecting differences due to communications delays). Ingredients, presences and unums communicate with each other via messages. When the trust boundaries are secured, ingredients can only communicate with other ingredients of the same presence, and presences can only communicate with other presences of the same unum (between agencies) and unums can only communicate with other unums in the same agency. These limitations are enforced by object references. If an object sends messages which violates a compact with a target object, the target object revokes its reference, thus preventing further messages from the offending sending object and thereby enforcing security rules.
According to the present invention, an unum is an object distributed over a plurality of agencies, which establishes an intra-unum/inter-agency communication trust boundary. Further according to the present invention, a presence is an agency-local instance of an unum inside this trust boundary. Further according to the present invention, an ingredient is a component part of an presence. Further according to the present invention, an agency is a logical process instantiated over a predetermined number of unums.
According to the present invention, an unum is established by creating ingredients at the level of its interface and attributes, and at the level of its implementation; and interconnecting ingredients into presences and unums.
According to the present invention, ingredients are able to communicate with other ingredients within a presence (on the same agency,) through an ingredient interface, as well as across an unum boundary through an unum interface.
Further according to the present invention, an ingredient communicates directly with a presence other than its own, within the same unum through a presence interface.
Further according to the present invention, an ingredient communicate directly with another unum within its own agency.
A further understanding of the nature and advantages of the inventions herein may be realized by reference to the remaining portions of the specification and the attached drawings.
Brief Description of the Drawings
FIG. 1 is a block diagram of a distributed computing system having a plurality of agencies according to the present invention.
FIG. 2 is a block diagram of the agencies shown in FIG. 1, but emphasizing different elements of the agencies, such as unums, presences, ingredients and various interfaces.
FIG. 3 is a block diagram illustrating the various interfaces of the distributed computing system shown in FIG. 2, including the various message that pass among elements of the system.
FIG. 4 is a block diagram showing several components of an ingredient.
FIG. 5 is a block diagram showing various components of a message as might be passed between components of the distributed computing system.
FIG. 6 is a block diagram of presence structure according to the present invention including role descriptors and method mappings.
FIG. 7 is a block diagram showing a generalized communication channel between two objects within a trust boundary and across a thrust boundary according to the present invention.
Description of the Preferred Embodiments
The preferred environment for the embodiments described here is a distributed world computing system. In the distributed world system, the world is partioned into regions, which can be analogized to rooms or geographical zones. The system is distributed over multiple agencies, which communicate with each other. As used herein, an agencies need not correspond with machine boundaries, as one agency could comprise multiple machines and one machine can run more than one agency. Agency boundaries correspond to trust boundaries, i.e., an agency "trusts" the code and data within its trust boundary and is suspicious of messages received from entities outside the agency's trust boundary. Whether the trust is misplaced is of only marginal relevance, because a well-designed and suspicious agency will not allow the corruption of its internals to the point where those internals are no longer trusted.
For each region, a number of agencies are "involved" with that region. One agency is the host agency for the region and the other agencies are participant agencies for the region. The region typically contains a number of unums which are associated with that region, and each agency which is involved with the region maintains a copy of each unum is the region, although the unums might be implemented differently on different agencies. Using the example of a virtual world, the region might be a room and the room might contain several avatar unums, a rock unum and a chair unum. As with the region, each unum has one host agency and zero or more participant agencies. The host for the unum controls the unum, but the host for the region also controls the unum. For example, the host for an avatar controls what the avatar will say and where it will move, but the region host controls whether the avatar will be able to move where it wants to move.
Host control is necessary to prevent a participant from "spoofing" an action with an unum, such as tricking all the other agencies into believing that an avatar's host just walked its avatar off a cliff, but also to limit the ill effects of a program which is running incorrectly. For example, if a region host determines that an unum host keeps trying to move the unum is a way that doesn't make any sense, the region host is able to remove the malfunctioning unum from the region.
An unum in a region has a "presence" in each agency involved with the region. The presences are dynamic, as agencies will come and go with their involvement with the region. The first presence of an unum is called the "prime presence" and is often the host presence (i.e., the presence which is on the host agency for that unum), but that need not always be the case, as a host may hand off control to a participant to be the new host for the unum. With multiple presences of an unum, there is always one host and one or more participants.
Each unum has a single, invariant class definition describing the unum (by its data structures and methods, etc.), but this definition need not be instantiated completely at every presence of the unum. Instead, each agency possesses a subset of the class definition suited to its platform and whether the agency is the host or a participant with respect to that unum. The class definition of the unum is made up of a collection of ingredients. Each ingredient is executable code, literal data, or a mixture of the two. When the class definition is instantiated as a presence, the instantiation also contains data structures containing state information for the unum. The state has two components: joint state and local state. Joint state data describes those attributes of the unum which are the semantically same at each presence (they might not be bit-for-bit the same and might be unsynchronized for short periods due to transmission delay). The local state is data that only the local presence cares about.
An agency may manipulate the local state of any unum that has a presence on the agency in any way the agency chooses. It can also alter its local copy of the joint state any way is chooses. However, this may result in the presence becoming desynchronized from the other presences of the unum and cause termination of the unum. If a participant is modifying its own copy of the joint state, it may do something that is impossible for the correct state and the participant agency may end up being ignored by the host agency for that unum and the host agency might inform all other participants to do the same. If the host agency modifies the joint state of the unum in a way that is not proper for that unum and broadcasts the new state to the participant agencies, some participant agencies might choose to remove the presence of the unum from its system. Since the misbehavior only affects the agency which is misbehaving, it doesn't matter. All that is required (and all that can actually be required) is that changes to the joint state be mediated through the host agency for the unum.
FIG. 1 shows two agencies, Agency X and Agency Y. The present invention can accommodate many more agencies, but only two are shown for simplicity. Agency X and Agency Y are mutually untrusting processes. This is not to say that any agency cannot be trusted, but just that the agencies are designed with the assumption that another agency may prove to be untrustworthy, either by design of the person or computer in control of the agency or by malfunction of a component of the agency.
Each agency includes a message I/O handler 100, a presence instantiator 102, various presences which have been instantiated by presence instantiator 102, a message compliance checker 106 and an object reference table 108. It should be understood that, in certain implementations, the various elements described herein might be just subcomponents or subfunctions of an integrated software system operating as the agency, or they may be separate components. For descriptive purposes, they are described here as being separable components. For example, the functionality of message compliance checker 106 might be distributed several ingredients and the agency itself.
The agencies communicate with interagency messages, which can be of different species. Some messages are messages from an agency itself to another agency, whereas other messages are from a presence on one agency to a presence on another agency. An agency also handles messages between presences on that agency (inter-unum messages). The agency might handle messages between ingredients which are wholly within a single presence (intrapresence messages), or these might be handled internal to the presence without routing through the agency's message I/O system.
FIG. 2 shows the two agencies of FIG. 1 in more detail. In FIG. 2, both agency X and agency Y have presences for unum R and unum S (presences RX, SX, RY and SY). Presence Rx is shown including two ingredients (ingredient A 122 and ingredient B 120). Presence RY is shown including an ingredient A 124 and an ingredient C 126. Presence SY is shown including an ingredient D 128. These ingredients are implementations of objects in an object-oriented programming sense. Ingredient A 122 corresponds to ingredient A 124, but they might not be identical if their respective agencies have different requirements for ingredients or different platforms. For example, if agency X is running on a Unix.TM. workstation with high-resolution graphics and agency Y is running on a Windows.RTM.-based personal computer with low-resolution graphics, ingredient A 122 might be written in C with high resolution display routines while ingredient A 124 is written in E with low resolution display routines.
Ingredients are somewhat independent of each other, in that one implementation of an ingredient can be substituted for another implementation, so long as both the implementation complies with the "contract" for that ingredient's "ingredient kind". The ingredient kind defines what interface the ingredient will present to other ingredients, presences within the ingredient's unum and unums within the ingredient's agency. Since ingredients are independent of each other, and there is no central security system controlling the use of ingredients, a mechanism is provided to control the behavior of ingredients in the distributed system.
The ingredients (and their logical aggreggations into presences and una) communicate by sending messages over interfaces. Several interfaces are shown in FIG. 2. Intrapresence interfaces 132, 140 control messages between two ingredients within a common presence. An interpresence interface 136 controls messages from an ingredient in one presence to a corresponding ingredient in another presence of the same unum. An inter-unum interface 144 controls messages from an ingredient in one presence on an agency to another unum (more precisely, the presence of that other unum on the same agency). Each interface spans a trust boundary and since, by definition, an object on the other side of a trust boundary cannot always be trusted to operate properly, the distributed system has mechanisms for blocking messages on any interface to limit the effects of a misbehaving ingredient.
FIG. 3 illustrates this point in more detail. As shown therein, when ingredient B 120 desires to communicate with ingredient A 122, it sends a message 130 over intrapresence interface 132. When ingredient A 122 desires to communicate with ingredient A 124, it sends a message 134 over interpresence interface 136. In many cases, one message will cause the generation of another message. For example if ingredient B 120 sends a message to ingredient A containing information which would change a state variable which is global over all presences of unum R, this will cause a message to be sent over interpresence interface 136 to ingredient A 124 (and all other ingredients A which are "neighbors" of ingredient A 122).
In order for messages to be routed to their target, messages include references to the target object. If a referenced object determines that a neighbor is not behaving properly (such as when the operator of an agency is running a hacked ingredient or the program code implementing the ingredient is flawed), the target reference can be revoked as to that neighbor. This is accomplished by the target changing its reference and informing all other, well-behaved neighbors of the new reference. This type of message control exists between intergredients, presences and una. Messages over interpresence interface 136 can be blocked by agency X or agency Y (since all messages ultimately pass through an agency-to-agency communications channel), or are blocked at an unum-agency boundary.
Other message paths shown in FIG. 3 include intrapresence interface 138 over which a message 138 is sent and inter-unum interface 144 over which message 142 is sent.
Referring now to FIG. 4, a basic structure of an ingredient 200 is shown. An ingredient has a kind, which defines its interface and how it will interact with other objects and has a list of neighbors. An ingredient also has elements found in conventional object-oriented objects, namely code to implement methods, constants and state variables (object data). The list of neighbors identifies the objects (ingredients, presences and una) to which the ingredient can send messages. If an ingredient receives messages indicating that another object has violated the terms of its contract (i.e., it is apparently not acting like its ingredient kind), that other object is deleted from the ingredient's list of neighbors and all other neighbors are provided with a new reference to that ingredient. Since messages are routed using the target reference, this effectively cuts off communication between the ingredient and the errant object.
An example of a message structure is shown in FIG. 5. A message 202 shown therein has a target reference 204, additional, optional) headers 206 for indicating the sender and including debugging information, and message data 212. Typically, the sender is not indicated in the headers of a message, so that messages can be transparently forwarded.
FIG. 6 is a block diagram of a presence implementation 250, which includes a list of method mappings 254, which describes the composition of a presence in terms of ingredients and encapsulates the concept of plumbing. More specifically, it describes the composition in terms of ingredient kinds. Each ingredient in the presence is assigned a role, which is simply a label that serves to identify that ingredient relative to the presence as a whole. A presence structure contains a method mapping relating the methods specified in the presence kind (and unum kind, as will be discussed below) to particular methods of component ingredients, enabling the ingredients to collectively realize the presence interface (and unum interface). These mappings define which methods of which ingredients handle which presence and unum interface functions. Referring back to FIG. 2, these mappings might indicate that messages to unum S are to be handled by a method in ingredient D. These mappings are necessary since the presences and una do not have methods of their own, but are just collections of ingredients. Of course, some ingredients' sole purpose might be to handle unum and presence intercommunication.
FIG. 7 is a block diagram illustrating how messages are sent over a trust boundary. Trust boundary 700 separates objects A and B from object C. This trust boundary can be an ingredient--ingredient, unum--unum, or agency--agency trust boundary. Since A and B are not separated by a trust boundary, object B trusts object A, so object A will always be able to send messages to object B. By contrast, object C does not always trust object A. To allow for communications which can be severed, a proxy for object C, C.sub.p, resides on the same side of trust boundary 700 as does object A. The proxy object's purpose is to send messages destined for object C across the trust boundary. At the other end, the message I/O subsystem 100 for the agency on which object C resides receives the message, reads its target reference and forwards it to the object specified in object reference table 108. A simplified example of the contents of object reference table 108 is shown in Table 1.
In a generalized messaging system, any message can be routed to any object by specifying an agency and an object in that agency's object reference table, if the object reference is a valid reference. Thus, a message from object A to object C will contain an indication of object C's agency and the object reference "235". Table 1 also shows that object reference 233 is a revoked reference. Any messages received by the agency with that object reference will be returned undelivered. Such messages might be expected when an object was misbehaving and continues to send messages after the offended object had its reference moved and only informed its well-behaved neighbors of the new object reference. Of course, in most practical implementations, the object references are numbers selected from a large enough pool that they cannot be guessed easily.
In the distributed object system described above, object-oriented objects are organized into ingredients and ingredients are logically organized into una distributed over agencies. When an agency first instantiates an unum, it creates a prime presence. As part of this process, the agency determines, from the unum kind, structure, etc., what the unum needs to operate (resources, nieghbors, etc.) and sets up the ingredients for the prime presence. This may involve retrieving code from prespecified locations to implement the methods of the ingredients.
When a second agency instantiates a nonprime presence of an existing unum, it must communicate with the prime presence (or another nonprime presence) so that the new presence is on the list of neighbors for one or more presences of the unum. The initialization of a nonprime presence is different from the instantiation of a prime presence, in that some assistance in setting up a new presence might be provided by the prime presence and in that the agency must perform the tasks of making sure that the presences are neighbors of each other so that intra-unum information can be shared. However, once instantiated, there is generally no continuing difference between a prime presence and a nonprime presence. Of course, in many cases the prime presence will have a host role and the nonprime presences will have participant roles with respect to particular state information. For example, an avatar unum might have a presence on the agency controlled by the person controlling the avatar. If the avatar unum has presences on other agencies, those presences look to the host presence to find out what the avatar is doing or wants to be doing. Nothing in the architecture, however, prevents a prime presence from transferring the host attribute to another presence.
The above description is illustrative and not restrictive. Many variations of the invention will become apparent to those of skill in the art upon review of this disclosure. The scope of the invention should, therefore, be determined not with reference to the above description, but instead should be determined with reference to the appended claims along with their full scope of equivalents. ##SPC1##
Claims
What is claimed is:
1. A data processing system comprising: at least first and second data processing presences, each of said at least first and second data processing presences being associated with a particular unum and each of said first and second data processing presences including a presence interface; at least a single unum distributed at least in part over said at least first and second data processing presences, each of said at least a single unum including an unum interface; and a plurality of ingredients associated with selected ones of said at least first and second data processing presences; each of said at least a single unum, each of said at least first and second data processing presences, and each of said plurality of ingredients, having an interface including attributes and methods, and each of said presence interfaces and said unum interfaces acting as a trust boundary.
2. The data processing system according to claim 1, wherein said plurality of ingredients includes non-interface ingredients and interface ingredients, each of said non-interface ingredients adapted for communication with others of said non-interface ingredients, and each of said interface ingredients being adapted for mapping messages across unums and data processing presences.
3. A distributed computing system comprising: a plurality of agencies, wherein an agency is a computing environment enclosed in an agency trust boundary within which computing resources and code are assumed by the enclosed agency to be reliable and trusted; a plurality of distributed objects, wherein each distributed object maintains state information and code for implementing the functionality of the distributed object and includes an interface for handling messages passing across an object boundary; a plurality of presences, wherein a presence is a container on a particular agency of the plurality of agencies for containing an instance of one or more of the distributed objects, each presence including a presence interface for handling messages passing across a presence boundary; a plurality of unums, wherein an unum is a collection of the presences of a particular distributed object; a messaging system that provides interface references to control message flow between ingredients within an agency boundary, between ingredients within a presence boundary and between presences within the unum boundary.
4. In a distributed computing system coupling distributed objects executing over a plurality of agencies, an agency being a computing platform secured within an agency trust boundary and interconnected to other agencies across the agency trust boundary, a method of securing execution of objects on agencies, wherein objects are not necessarily trusted by agencies that execute those objects and agencies are not necessarily trusted by the objects executed on those agencies, a method of securing mutually controllable execution of the objects on the agencies comprising the steps of: instantiating a plurality of instances of ingredients, wherein an instance of an ingredient is an executable object containing state information that maintains a state of the instance, method code that encode for the functions performed by the instance and an instance interface for passing messages to and from the instance; associating each instance with an agency, wherein the associated agency is the computing platform on which the instance is executed; associating each instance with a presence, wherein a presence is a logical association of mutually trusted instances on one agency; associating each presence with a unum, a unum being a logical association of presences associating instances of common ingredients, wherein each instance of a common ingredient is instantiated on an agency associated with a presence of the unum; and controlling message flow between instances of ingredients such that a message from a first instance of a first ingredient executing on a first agency to an instance of a second ingredient executing on a second agency passes through at least an ingredient interface between the first instance of the first ingredient and a second instance of the first ingredient, wherein the second instance of the first ingredient is executing on the second agency the first and second instances are associated with a common unum, and an interface between the second instance of the first ingredient and the instance of the second ingredient, wherein that interface is an intrapresence interface when the first and second ingredients share a common presence on the second agency and that interface is an interpresence interface when the first and second ingredients do not share a common presence on the second agency.
5. The method of claim 4, wherein the step of controlling message flow comprising the steps of: limiting messages to an instance of an ingredient to messages that contain a reference to that instance; and limiting messages to a presence to messages that contain a reference to that presence, wherein the references are unforgeable references, an unforgeable reference being a reference to an object that cannot be generated without an authorization originated from a creator of the object using less than a threshold amount of computing effort.
6. The method of claim 4, wherein the target recipient of a message from a sending ingredient within a presence in an agency is one of an ingredient within the sending ingredient's presence, an unum within the agency, or a presence on another agency but within the sending ingredient's unum.
7. The method of claim 4, wherein messages to an unum target recipient comply with an unum interface, messages to a presence target recipient comply with a presence interface, and messages to an ingredient target recipient comply with an ingredient interface.
8. The method of claim 5, wherein said reference is a revocable reference, the method further comprising: determining if a message passed to an interface is a nonconforming message in that the message fails to conform to a predetermined set of rules for the interface; and revoking the reference to an object to which the nonconforming message is directed.
Patent Citations (5)
Non-Patent Literature (2)
- (no author given) ORB 2.0 RFP Submission, OMG TC Document 94.9.32, pp. 1-109, Sep. 1994.
- (no author given) "The Common Object Rquest Broker: Architecture and Specification" Digital Equipment Corporation, OMG Document No. 91.12.1 Revision 1.1, pp. 1-177, 1992.