A FRAMEWORK FOR DISTRIBUTED OBJECTS IN PEER-TO-PEER

COOPERATION ENVIRONMENTS

Bernd Eßmann, Thorsten Hampel

Heinz Nixdorf Institute

University of Paderborn

urstenallee 11, 31102 Paderborn

Keywords:

Mobile Computing, Spontaneous Collaboration, Distributed Knowledge Spaces, CSCW.

Abstract:

Mobile forms of cooperative knowledge organization need system architectures that also allow spontaneous

(ad hoc) networking and collaboration structures. An essential requirement here, besides establishing the peer-

to-peer network, is the design of suitable framework architectures for object distribution. This article presents

our approach to developing a basic architecture for distributed systems that support cooperation. This also

involves creating interfaces to existing classical CSCW architectures and systems. The novel element of our

approach, which is based on a JXTA network, is its use of JXTA services to distribute objects among peers,

thus achieving the desired object distribution.

1 INTRODUCTION

Mobility is one of the most important of modern-

day developments. Methods, tools and architectures

for CSCW systems must take into account mobile,

ﬂexible forms of cooperation. Cooperation partners

and learners spontaneously form groups, making use

of technologies for spontaneous networking. At the

same time, it is essential to take advantage of today’s

powerful mobile devices, such as PDAs, smartphones

as well as laptops, to develop mechanisms that sup-

port interpersonal collaboration without existing ﬁxed

infrastructures. And it is also important to suitably

integrate classical CSCW systems into such mobile

forms of cooperation, where they are available.

One possible way to provide ubiquitous network

environments without the need for centralized ad-

ministration is to establish so-called ad hoc networks

(Perkins, 2001). Ad hoc networks appear to be the in-

frastructure of choice for network support in mobile

environments where no guaranteed connections ex-

ist. One of the main features of such networks is the

ability to establish network structures spontaneously

from scratch, though they are not reliable. The dy-

namic nature of ad hoc networks means that hosts

may appear or disappear without warning; even com-

plete networks may be disconnected (network parti-

tioning) (Feeney et al., 2001).

Applications running on such networks need to be

aware of the loss of connection to counterparts in the

network. Strategies for network failures are espe-

cially important for systems providing CSCW envi-

ronments.

Our approach to the challenges of mobile CSCW

environments is to provide a basic architecture sup-

porting small highly portable devices in ad hoc net-

works as well as powerful CSCW servers in existing

network infrastructures. Major services such as net-

work communication, a distributed object repository

and user interfaces should be integrated in a ﬂexible

and modular manner. This will allow extension of the

basic architecture in future developments – an impor-

tant feature since many of the above-mentioned prob-

lems are the subject of current research activities.

As a ﬁrst step toward designing CSCW systems

that suitably integrate peer-to-peer and client-server-

based architectural concepts, we begin by presenting

a basic object distribution and administration mech-

anism. This allows the dynamic transmission of ob-

jects between ad hoc networked peers and the testing

of ﬂexible strategies for replication and object distrib-

ution. The approach involves ﬂexibly integrating clas-

sical client-server-based CSCW architectures as so-

called super-peers – depending on their availability.

The paper is organized as follows: First we intro-

duce systems that address some of the requirements

for mobile working environments. Then we look at

the fundamental problems of object management in

157

Eßmann B. and Hampel T. (2005).

A FRAMEWORK FOR DISTRIBUTED OBJECTS IN PEER-TO-PEER COOPERATION ENVIRONMENTS.

In Proceedings of the Seventh International Conference on Enterprise Information Systems, pages 157-162

DOI: 10.5220/0002528601570162

 SciTePress

mobile CSCW environments. The following two sec-

tions present our conceptual approach and give a brief

description of our framework’s architecture. The pa-

per ends with a brief conclusion and a look at future

research prospects.

2 RELATED WORK

Peer-to-peer platforms are ﬂexible depending in terms

of the network infrastructure. Most peer-to-peer solu-

tions are designed to distribute data, communication

or computing resources over the network. Neverthe-

less, even when they are called peer-to-peer, many

of these platforms still use certain client-server struc-

tures for indexing the distributed data, for connection

management and other functions. Since they require

Internet access, most of these tools do not allow data

exchange in a dynamic ad hoc network. (Milojicic

et al., 2002) give an overview of existing peer-to-peer

systems.

A well-known peer-to-peer system designed for co-

operative work is Groove

. Developed as a peer-to

peer CSCW platform, it features distributed persis-

tence. In view of the architecture for synchroniz-

ing the distributed workspaces, Groove still requires

servers.

For communication in heterogeneous network en-

vironments with multiple devices and applications,

the Park Labs developed Speakeasy (Edwards et al.,

2002b). They call their approach recombinant

computing. This means providing ﬁxed domain-

independent interfaces and mobile code. Speakeasy

is designed to interconnect appliances such as PDAs

and multimedia devices in the environment on a peer-

to-peer basis. Casca is an application that uses the

Speakeasy technology for cooperative work (Edwards

et al., 2002a). It allows the detection and selection of

potential partners for collaborative work. Users can

share documents in shared spaces. These spaces may

also contain deviceslike printers and beamers but they

do not provide semantic or object-oriented structures

like virtual knowledge spaces.

3 OBJECT DISTRIBUTION

Client-server architectures are no longer useful in dy-

namic network infrastructures. The dedicated server

is a single point of failure. When the server is dis-

connected from the network, all clients are rendered

useless. Figure 1 shows objects stored in a central

repository in the network, as is common in client-

server architectures. Clients access original objects

http://www.groove.net

...

Figure 1: Objects stored in a centralized object repository.

Clients reference the original objects by proxy objects.

through proxy objects. These are merely pointers to

the original objects stored in the central server and its

object repository. When disconnected from the server,

clients are no longer able to access the objects.

The way to avoid these shortcomings of ad hoc net-

works is to use peer-to-peer architectures. They are

designed to distribute the working environment over

all involved devices. This strategy avoids a complete

breakdown of the working environment as a result of

network disconnections and ensures that resources on

all connected devices remain available – the ﬁrst key

to supporting mobility.

There are several strategies for distributing objects

over peer-to-peer networks. To get some idea of the

strategies available, it might be useful to look at the

various stages of a connected distributed object repos-

itory:

1. First, each peer is disconnected from the network,

only retaining access to its own local repository.

The user working on such a peer may only use these

local objects (Figure 2a).

2. During connection to the network, the peer will

hopefully discover some counterparts, which will

themselves provide objects in their local object

repositories. Now the peers can also access the re-

mote objects via the peer-to-peer network layer. To

enable the remote objects to be manipulated trans-

parently to users, they are represented by so-called

proxy objects (Figure 2b).

3. When disconnected from a node hosting a remote

object, the proxy objects point to a no longer exist-

ing object. In this state, the peer’s object structure

is inconsistent (Figure 2c).

Most systems providing a distributed object repos-

itory try several strategies to deal with disconnection

ICEIS 2005 - SOFTWARE AGENTS AND INTERNET COMPUTING

158

a) unconnected peers with local objects

b) connected peers with distributed objects

c) disconnected peers with inconsistent objects

d) disconnected peers with replicated objects

Figure 2: A peer-to-peer-based distributed object reposi-

tory. Every peer may reference remote objects and provide

objects to foreign peers.

from peers hosting remote objects. Often, they repli-

cate the objects over the network. Replication may

be done using several strategies. File-sharing applica-

tions, for instance, replicate the objects to every peer

requesting an object. Other systems use heuristics to

distribute the data over a subset of the available peers.

To achieve better distribution, it is also possible to dis-

tribute chunks of the data, instead of the complete ﬁle,

over the peers. The heuristics used often include sta-

tistics about the online time and storage capacity for

selection of the peers to save the objects.

File-sharing systems simply distribute objects

without enabling them to be changed once they are

stored in the network. In CSCW environments, it

is vital to be able to change the objects for cooper-

ation purposes. In the simplest replication strategy,

replicated objects can be seen as proxy objects with

an embodied copy of the remote object. When con-

nected to the peer hosting the original objects, they

point to the remote objects. In the case of disconnec-

tion, the replicated objects merely change the pointer

to themselves; the user can work on the local replica

(see Figure 2d). As soon as they are reconnected, the

replicas of an object must be synchronized with the

original and the proxy objects point to the originals

again. Thus, when working on replicated objects, it is

important to synchronize the distributed instances of

an object. Concurrent changes on instances could put

the distributed object in an inconsistent state.

In our approach, we have not yet tackled the prob-

lems of distributed object repositories, preferring to

ﬁrst establish a peer-to-peer communication network

to connect peers in heterogeneous networks transpar-

ently to the user. The important thing here is that users

need not know the network environment they are con-

nected to in order to participate in the cooperation en-

vironment.

To establish the network communication, we use

JXTA (Gong, 2001), an industry-standard peer-to-

peer framework. JXTA allows peer-to-peer communi-

cation and service detection even in ﬁrewall- or NAT-

protected networks. Aiming at an object-oriented,

distributed cooperation environment, our implemen-

tation transforms the service-oriented philosophy of

JXTA into our philosophy of ﬂexible object distribu-

tion.

Future versions of our framework will use the ﬂex-

ibility of the framework’s design to integrate different

strategies of object distribution and replication. At

present, however, the objects are stored locally on a

single peer, allowing other peers to access them re-

motely. When disconnected, the peers lose access

to the remote objects. To reduce the risk of dis-

connection from transient peers, we included the op-

tion of integrating dedicated CSCW servers into the

peer-to-peer cooperation environment. For this pur-

pose, a peer becomes a proxy to the classical client-

server CSCW server. Peers can thus access the objects

stored on the server, even though the server knows

nothing about the peer-to-peer network. Users of

peers can store their cooperation results from within

the peer-to-peer environment on dedicated CSCW

servers. This scenario is shown in Figure 3.

4 OBJECTS, TYPE CLASSES AND

VIEWS

The distributed cooperation environment is based on

the objects living in the distributed knowledge spaces.

Peers building this environment run in a heteroge-

neous hardware and software environment, which

means that many different implementations of the

peers may exist. It is not a proper solution to update

and restart all peers when new object types are intro-

duced or existing ones are altered.

To tackle this problem, we provide type classes car-

rying the functionality for manipulating and creating

objects of a certain type. Each object type is related to

A FRAMEWORK FOR DISTRIBUTED OBJECTS IN PEER-TO-PEER COOPERATION ENVIRONMENTS

159

proxy peer

CSCW server

proxy peer

CSCW

server

proxy objects for

server repository

centralized

repository

Figure 3: A peer-to-peer-based distributed object repository

with a proxy peer acting as a gateway to a dedicated CSCW

server.

exactly one type class. To distribute the type classes to

the peer, they are stored in the same repository as the

objects. New versions of a type class overwrite older

ones as soon as they appear in the repository. When

a peer does not know how to handle a new type of

object, it must simply load the appropriate type class.

The concept for viewing an object’s attributes and

content is similar. Again, the objects of a speciﬁc

type are related to so-called views. Unlike the type

classes, each object type may be related to several

views and users can choose the view that best matches

their needs.

If participants in the peer-to-peer cooperation en-

vironment wish to provide a new object type, they

must supply a type-class object, describing the ob-

ject type’s structure and functionality, and at least

one view to show its content and attributes within the

peer’s graphical user interface (GUI). Figure 4 shows

an object type with its type class and view.

An object type can be introduced or altered at run-

time without restarting any peers in the network. In-

deed, views and management code are optional. Al-

though provision of an adapted type class and view

is recommended, all objects can be browsed and dis-

played using default attributes; type-dependent values

can be ignored. The additional attributes featured by

the type class merely provide further information use-

ful for the associated object type.

An object’s standard attributes include a reference

list of associated objects. These references to other

Core

GUI

Typeclass

--------------

file.swifffClass

- create content body

- object functions

Object

--------------

type="file"

View

--------------

file.swifffView

- visualization

- user interaction

Figure 4: Every object type may provide some views and

code for manipulation.

objects cause a graph-like object structure to evolve,

which builds the semantics of the distributed knowl-

edge space. The system does not check any con-

straints to on the references because they are provided

by the users’ actions. Again, this approach is highly

ﬂexible in terms of programming and expanding func-

tionality in the future.

5 DISTRIBUTED KNOWLEDGE

SPACES

For seamless cooperation in mobile scenarios, the co-

operation environment must be independent of ex-

isting infrastructures. This is also true of the net-

work environment and the organizational structures.

Important requirements for the cooperation environ-

ment are self-administration and the option of unre-

stricted structuring of knowledge. For several years

now, cooperative knowledge spaces have been the

main conceptual goal in our efforts to establish collab-

orative structures between cooperation partners. Vir-

tual knowledge spaces allow cooperation partners to

establish groups and associated areas for their co-

operation. Different cooperation areas may be con-

nected by gates, allowing the user – represented by

a virtual avatar – to move from one area to another.

Objects and groups in virtual knowledge spaces are

persistent, enabling them to be used in later cooper-

ation sessions. The main idea behind the knowledge

space metaphor is that all the different services, doc-

uments and objects can be interreferenced in an in-

tegrated manner. To transpose this concept to peer-

to-peer environments, where objects are distributed

over the network, the concept of virtual knowledge

spaces must be enhanced. We call this new notion

of knowledge spaces spread over the network distrib-

uted knowledge spaces. The cooperation area is saved

in its entirety on a device and can be replicated by col-

laborators. Gates are marked as nonfunctional if the

target area is not available.

For short-term cooperation, we have presented the

concept of temporary groups and knowledge areas

(Eßmann and Hampel, 2003). They are particularly

ICEIS 2005 - SOFTWARE AGENTS AND INTERNET COMPUTING

160

useful in spontaneous face-to-face situations where

fast establishment of a cooperation setup is needed.

Once the cooperation session is closed, the group and

the area are deleted. Optionally, they may be included

in persistent cooperation structures.

While the above concepts allow seamless coopera-

tion in mobile environments without functional con-

straints, systems implementing these concepts must

take into account different network structures. Ba-

sic requirements for an application implementing a

mobility-supporting cooperation system are open ar-

chitecture, automatic conﬁguration and spontaneous

networking (Eßmann et al., 2004a). The ﬁrst step was

to implement a basic framework with simple but pow-

erful modules to support mobile cooperation. This ar-

chitecture is described in the next section.

6 FRAMEWORK

ARCHITECTURE

The objective of creating a mobility-aware coopera-

tion platform thus involves implementing the concept

of distributed knowledge spaces. The technology is

based on a peer-to-peer architecture supporting mul-

tiple types of static and mobile devices. The appli-

cation must, then, be platform-independent and use

standardized interfaces and protocols.

To achieve maximum ﬂexibility in terms of provid-

ing different device-speciﬁc user interfaces, our pro-

totype implementation is split into a core and a user

interface part. The communication interfaces are im-

plemented as modules, which can be extended or ex-

changed. The graphical user interfaces visualize the

objects in the knowledge space maintained by the

core. While each core instance represents exactly one

user of the cooperation environment, it is possible to

connect more than one user interface to the core. Ad-

ditionally, it is possible to run a core instance with-

out a user interface. From the user interface point of

view, it may run on a mobile device without a local

core instance by using a remote one. Figure 5 shows

a possible user scenario, where each user interface is

connected to exactly one core, but using several or no

connected user interfaces.

While the core is responsible for persistence and

object management, the modules are mainly responsi-

ble for the network communication. To ensure adapt-

ability to future changes, the design of the core is sim-

ilar to that of microkernel architectures, allowing indi-

vidual parts to be exchanged if needed. For a detailed

description of the core’s implementation in our proto-

type, please refer to (Eßmann et al., 2004b). Further

information on the user interface concepts and imple-

mentation can be found in (Slawik et al., 2004).

We now go on to present architectural parts vital

JXTA network

core

GUI GUI

RMI

core

RMI

GUI

core

simple GUI

RMI

Figure 5: Peer-to-peer-connected cores, each assigned to

exactly one user. Optionally, the core can be used by one or

more user interfaces.

to our framework. Given the ﬂexible concept of ob-

jects, type classes and views presented in Section 4,

new object types can be added to the system without

affecting the core’s implementation. An important el-

ement in this concept is a ﬂexible object loader, in-

cluding a persistence module. While the object loader

manages all objects handled by the application, the

persistence module manages access to the local data

repository. The job of the loader/persistence module

is to manage the remote/local storage of objects, re-

spectively. Here, a replication scheme would plug

in. Local objects are accessed via a repository con-

troller from the local repository; remote objects are

accessed via the network interface module from the

remote peer, which hosts the respective object. This

process is transparent to the peer itself: it does not

have to take into consideration where the objects are

located.

An access control module checks the users’ permis-

sions concerning a requested object. For this purpose,

objects provide an access control list (ACL). The ac-

cess rights are checked on every access to an object.

External communication is provided via the GUI

interface and network interface modules. The GUI

interface provides access for one or more user inter-

faces, while the network interface is responsible for

all network communication between the peers. In our

prototype, this is an interface to the JXTA network

but it may be replaced or supplemented by a differ-

ent communication layer in the future. We use the

network layer abstraction of JXTA to obtain adequate

peer-to-peer communication. A discovery manager

is closely coupled with the JXTA interface and han-

dles service discovery. If new potential cooperation

partners appear, they are instantly made known in the

knowledge space by the creation of a new user object.

All known users are monitored according to their on-

line status.

Internal object communication is based on XML.

Thus, a parser module provides XSLT as a query lan-

guage to receive and ﬁlter objects. This feature is used

for minimizing communication overhead. The user

interfaces only receive the object properties they re-

A FRAMEWORK FOR DISTRIBUTED OBJECTS IN PEER-TO-PEER COOPERATION ENVIRONMENTS

161

ally need to present to the user. On the assumption

that data are more often read than written, this form

of optimization is only done for read access.

These central modules enable the whole function-

ality of the framework to be maintained. While the

object loader, persistence module and access control

handle internal object access, the (JXTA) network in-

terface and the GUI interface allow transparent com-

munication with external parts of the peer’s core.

7 CONCLUSION

Given the wide range of relevant questions pertaining

to mobile (e.g. peer-to-peer-based) forms of coopera-

tive work, the ﬁrst requirement here is the creation of

an open and ﬂexible framework that allows us both

to study different research issues relating to object

distribution and replication strategies and to develop

and test practicable systems in conjunction with ex-

isting classical CSCW systems. Another important

aspect is the separation of user interface and system

core in a network-transparent manner. Devices that

are unable to compute the whole core’s functionality

can utilize other network-connected devices to pro-

vide the required core. The system is thus scalable

from low-end devices like smart-phones to powerful

devices like high-end laptops or even workstations

and servers. The ﬂexible object design includes – but

is not limited to – user interface design. Our concept

allows developers to provide new object types with

related views and functionality at runtime. A node

that does not know how to handle an object type can

load the views and functionality from other peers. By

using Java, XML and JXTA, we support a wide range

of devices and operating systems. This also ensures

that our solution is based on open interfaces and stan-

dards. Of course, later versions of our system may

exchange or extend the communication protocols be-

cause of our open module architecture. Although it

fails to provide a comprehensive functionality, our ap-

proach offers a mobility-supporting cooperative envi-

ronment based on open standards and open interfaces.

Our framework is therefore a ﬁrst step toward the de-

velopment of an open peer-to-peer architecture for

distributed knowledge spaces. With its object distrib-

ution mechanisms and support for ad hoc networks, it

allows research into new forms of mobility in cooper-

ative knowledge management.

ACKNOWLEDGMENTS

Bernd Eßmann is a member of the postgraduate pro-

gram 776 ”Automatic Conﬁguration in Open Sys-

tems”, which is funded by the German Research

Foundation (DFG) and the Heinz Nixdorf Institute.

REFERENCES

Edwards, W. K., Newman, M. W., Sedivy, J. Z., Smith, T. F.,

Balfanz, D., Smetters, D. K., Wong, H. C., and Izadi,

S. (2002a). Using speakeasy for ad hoc peer-to-peer

collaboration. In conference on Computer Supported

Cooperative Work (CSCW’02), pages 256–265, New

Orlando, Louisiana, USA. ACM Press, New York,

NY, USA.

Edwards, W. K., Newman, M. W., Sedivy, J. Z., Smith,

T. F., and Izadi, S. (2002b). Challenge: Recombi-

nant computing and the speakeasy approach. In Inter-

national Conference on Mobile Computing and Net-

working (MOBICOM’02), pages 279–286, Atlanta,

Georgia, USA. ACM Press, New York, NY, USA.

Eßmann, B. and Hampel, T. (2003). Integrating coopera-

tive knowledge spaces into mobile environments. In

Rossett, A., editor, E-Learn 2003, pages 225–232,

Phoenix, AZ, USA.

Eßmann, B., Hampel, T., and Bopp, T. (2004a). A net-

work component architecture for collaboration in mo-

bile settings. In Seruca, I., Filipe, J., Hammoudi, S.,

and Cordeiro, J., editors, 6th International Conference

on Enterprise Information Systems 2004 (ICEIS’04),

volume 4, pages 337–343, Porto, Portugal.

Eßmann, B., Hampel, T., and Slawik, J. (2004b). A jxta-

based framework for mobile cooperation in distributed

knowledge spaces. In Karagianis, D. and Reimer, U.,

editors, 5th International Conference on Practical As-

pects of Knowledge Management (PAKM’04), Lecture

Notes in Artiﬁcial Intelligence, pages 11–22, Vienna,

Austria. Springer Verlag.

Feeney, L. M., Ahlgren, B., and Westerlund, A. (2001).

Spontaneous networking: An application-oriented ap-

proach to ad hoc networking. IEEE Communications

Magazin, 39:176–181.

Gong, L. (2001). Jxta: A network programming environ-

ment. IEEE Internet Computing, 5:88–95.

Milojicic, D. S., Kalogeraki, V., Lukose, R., Nagaraja, K.,

Pruyne, J., Richard, B., Rollins, S., and Xu, Z. (2002).

Peer-to-peer computing. Technical report, HP Labs.

Perkins, C. E., editor (2001). Ad Hoc Networking. Addison

Wesley, Boston, USA.

Slawik, J., Eßmann, B., and Hampel, T. (2004). Shared

views on mobile knowledge – a concept of a graphi-

cal user interface. In Karagianis, D. and Reimer, U.,

editors, 5th International Conference on Practical As-

pects of Knowledge Management (PAKM’04), Lecture

Notes in Artiﬁcial Intelligence, pages 82–93, Vienna,

Austria. Springer Verlag.

ICEIS 2005 - SOFTWARE AGENTS AND INTERNET COMPUTING

162