In this paper, we propose context-based peer clustering approaches for point-of-interest (POI) prediction using peer-to-peer federated learning (P2P-FL). Our approach addresses the challenges of non-iid data and single-server vulnerabilities commonly encountered in federated learning (FL) environments. By removing reliance on a central aggregator, P2P-FL avoids single points of failure or trust, promoting fully decentralized collaboration among peers. FL and P2P-FL approaches struggle with data heterogeneity, resulting in suboptimal model performance. To overcome this, we focus on context-based peer clustering rather than random peer selection to improve model convergence and accuracy by allowing peers with shared cultural or temporal preferences to exchange more relevant model parameters. This context-based peer clustering improves model convergence and prediction accuracy by enabling peers to exchange more relevant model parameters. To evaluate our solution, we conduct experiments on a real-world dataset, comparing the performance of our context-based clustering against a random peer selection baseline. The results demonstrate that context-based peer clustering improves the POI prediction quality. Additionally, we analyze trade-offs between different criteria. This work highlights the potential of P2P-FL to safeguard peer privacy while delivering personalized POI predictions. Furthermore, it broadens the scope of FL-based applications, emphasizing the importance of context-based collaboration in P2P-FL solutions.

Due to the limitations of existing Global Navigation and Satellite Systems in non-line-of-sight environments, Indoor Tracking Systems (ITS) have been extensively discussed in the literature. These systems are split into two main categories: infrastructure-based and infrastructure-less. For infrastructure-based ITS, the challenge is the cost of deploying and maintaining infrastructure. For infrastructure-less ITS, mostly based on inertial sensors, the challenge is the accumulation of errors due to noisy sensors. This paper proposes a collaborative ITS that balances accuracy and cost to address these challenges. Our approach combines three techniques. First, each mobile device estimates its location using a Pedestrian Dead Reckoning (PDR) algorithm. Second, we rely on a collaborative scheme where mobile devices exchange their estimated locations when close to each other to correct those estimates. Third, we deploy a limited number of fixed Bluetooth Low-Energy (BLE) beacons to correct those estimates further, using a spatial indexing technique to identify strategic locations for placing the beacons. We show that collaboration between mobile devices coupled with only two beacons, with a density of 1 beacon every $4200 \mathrm{~m}^{2}$, increases the accuracy of a baseline PDR algorithm by up to 23%. With four beacons, with a density of 1 beacon for every $2100 \mathrm{~m}^{2}$, our solution offers an accuracy of 5.7 meters, corresponding to a 52% accuracy improvement compared to the PDR.

Internet users possess accounts on dozens of online services where they are often identified by one of their e-mail addresses. They often use the same address on multiple services and for communicating with their contacts. In this paper, we investigate attacks that enable an adversary (e.g., company, friend) to determine (stealthily or not) whether an individual, identified by their e-mail address, has an account on certain services (i.e., an account enumeration attack ). Such attacks on account privacy have serious implications as information about one’s accounts can be used to (1) profile them and (2) improve the effectiveness of phishing. We take a multifaceted approach and study these attacks through a combination of experiments (63 services), surveys (318 respondents), and focus groups (13 participants). We demonstrate the high vulnerability of popular services (93.7%) and the concerns of users about their account privacy, as well as their increased susceptibility to phishing e-mails that impersonate services on which they have an account. We also provide findings on the challenges in implementing countermeasures for service providers and on users’ ideas for enhancing their account privacy. Finally, our interaction with national data protection authorities led to the inclusion of recommendations in their developers’ guide.

In this paper, we present the insights drawn from a comprehensive survey of human-mobility modeling research based on 1680 articles that can serve as a roadmap for research and practice in this area. Mobility modeling research has accelerated the advancement of several fields of studies such as urban planning, epidemic modeling, traffic engineering and contributed to the development of location-based services. However, while the application of mobility models in different domains has increased, the credibility of the research results has decreased. We highlight two significant shortfalls observed in our reviewed studies: (1) data-agnostic model selection resulting in a poor tradeoff between modeling accuracy vs. computational complexity, and (2) failure in identifying the source of empirical gains, due to adoption of erroneous validation methodologies. We also observe troubling trends with respect to the application of Markov models for modeling mobility, despite the questionable association between Markov processes and mobility dynamics. We offer the literature meta-data and the associated tools to the community in order to improve the reliability and credibility of human mobility modeling research.

Rich human mobility datasets are fundamental for evaluating algorithms pertaining to geographic information systems. Unfortunately, existing mobility datasets--that are available to the research community--are restricted to location data captured through a single sensor (typically GPS) and have a low spatiotemporal granularity. They also lack ground-truth data regarding points of interest and the associated semantic labels (e.g., "home", "work", etc.). In this paper, we present Breadcrumbs, a rich mobility dataset collected from multiple sensors (incl. GPS, GSM, WiFi, Bluetooth) on the smartphones of 81 individuals. In addition to sensor data, Breadcrumbs contains ground-truth data regarding people points of interest (incl. semantic labels) as well as demographic attributes, contact records, calendar events, lifestyle information, and social relationship labels between the participants of the study. We describe the data collection methodology and present a preliminary quantitative analysis of the dataset. A sanitized version of the dataset as well as the source code will be made available to the research community.

Finding trajectories and discovering motifs that are similar in large datasets is a central problem for a wide range of applications. Solutions addressing this problem usually rely on spatial indexing and on the computation of a similarity measure in polynomial time. Although effective in the context of sparse trajectory datasets, this approach is too expensive in the context of dense datasets, where many trajectories potentially match with a given query. In this paper, we apply fingerprinting, a copy-detection mechanism used in the context of textual data, to trajectories. To this end, we fingerprint trajectories with geodabs, a construction based on geohash aimed at trajectory fingerprinting. We demonstrate that by relying on the properties of a space filling curve geodabs can be used to build sharded inverted indexes. We show how normalization affects precision and recall, two key measures in information retrieval. We then demonstrate that the probabilistic nature of fingerprinting has a marginal effect on the quality of the results. Finally, we evaluate our method in terms of performances and show that, in contrast with existing methods, it is not affected by the density of the trajectory dataset and that it can be efficiently distributed.

The proliferation of mobile devices equipped with internet connectivity and global positioning functionality (GPS) has resulted in the generation of large volumes of spatiotemporal data. This has led to the rapid evolution of location-based services. The anticipatory nature of these services, demand exploitation of a broader range of user information for service personalization. Determining the users' places of interest, i.e. hotspots is critical to understand their behaviors and preferences. Existing techniques to detect hotspots rely on a set of a-priori determined parameters that are either dataset dependent or derived without any empirical basis. This leads to biased results and inaccuracies in estimating the total number of hotspots belonging to a user, their shape and the average dwelling time. In this paper, we propose a parameter-less technique for extracting hotspots from spatiotemporal trajectories without any a-priori assumptions. We eliminate parameter dependence by treating trajectories as spatiotemporal signals and rely on signal processing algorithms to derive hotspots. We experimentally show that, our technique does not necessitate any spatiotemporal or behavior dependent bounds, which makes it suitable to extract hotspots from a larger variety of datasets and across users having disparate mobility behaviors. Our evaluation results on a real world dataset, show accuracy rates exceeding 80% and outperforms traditional clustering techniques used for hotspot detection.

The rise of the Internet of things (IoT) poses massive scalability issues for location-based services. More particularly, location-aware publish and subscribe services are struggling to scale out the computation of matches between publications and subscriptions that continuously update their location. In this demonstration paper, we propose a novel distributed and horizontally scalable architecture for location-aware publish and subscribe. Our middleware architecture relies on a multi-step routing mechanism based on consistent hashing and range partitioning. To demonstrate its scalability, we present a traffic data generator, which, in contrast to existing generators, can be used to perform real-time load tests. Finally, we show that our architecture can be deployed on a small 10-node cluster and can process up to 80,000 location updates per second producing 25,000 matches per seconds.

With billions of connected users and objects, location-based services face a massive scalability challenge. We propose a horizontally-scalable and reliable location-based publish/subscribe architecture that can be deployed on a cluster made of commodity hardware. As many modern location-based publish/subscribe systems, our architecture supports moving publishers, as well as moving subscribers. When a publication moves in the range of a subscription, the owner of this subscription is instantly notified via a server-initiated event, usually in the form of a push notification. To achieve this, most existing solutions rely on classic indexing data structures, such as R-trees, and they struggle at scaling beyond the scope of a single computing unit. Our architecture introduces a multi-step routing mechanism that, to achieve horizontal scalability, efficiently combines range partitioning, consistent hashing and a min-wise hashing agreement. In case of node failure, an active replication strategy ensures a reliable delivery of publication throughout the multistep routing mechanism. From an algorithmic perspective, we show that the number of messages required to compute a match is optimal in the execution model we consider and that the number of routing steps is constant. Using experimental results, we show that our method achieves high throughput, low latency and scales horizontally. For example, with a cluster made of 200~nodes, our architecture can process up to 190'000 location updates per second for a fleet of nearly 1'900'000 moving entities, producing more than 130'000 matches per second.

Location-based services today, exceedingly depend on user mobility prediction, in order to push context aware services ahead of time. Existing location forecasting techniques are driven by large volumes of data to train the prediction models in a centralised server. This amounts to considerably long waiting times before the model kicks in. Disclosing highly sensitive location information to third party entities also exposes the user to several privacy risks. To address these issues, we put forth a mobility prediction system, able to provide swift realtime predictions, evading the strenuous training procedure. We enable this by constantly adapting the model to substantive user mobility behaviours that facilitate accurate predictions even on marginal time bounded movements. In comparison to existing frameworks, we utilise less volumes of data to produce satisfactory prediction accuracies. This in turn lowers the computational complexity making implementation on mobile devices feasible and a step towards privacy preservation. Here, only the predicted location can be sent to such services to maintain the utility/privacy tradeoff. Our preliminary evaluations based on real world mobility traces corroborate our hypothesis.

We introduce an algorithm that implements a time-limited neighbor detector service in mobile ad hoc networks. The time-limited neighbor detector enables a mobile device to detect nearby devices in the past, present and up to some bounded time interval in the future. In particular, it can be used by a new trend of mobile applications known as proximity-based mobile applications. To implement the neighbor detector, our algorithm uses n = 2kvirtual mobile nodes where k is a non-negative integer. A virtual mobile node is an abstraction that is akin to a mobile node that travels in the network in a predefined trajectory. In practice, it can be implemented by a set of mobile nodes based on a replicated state machine approach. Our algorithm implements the neighbor detector for nodes located in a circular region. We assume that each node can accurately predict its own locations up to some bounded time interval Δpredictin the future. The key idea of the algorithm is that the virtual mobile nodes regularly collect location predictions of nodes from different subregions, meet to share what they have collected with each other and then distribute the collected location predictions to nodes. Thus, each node can use the distributed location predictions for neighbor detection. We show that our algorithm is correct under certain conditions. Compared to a solution that works with a single virtual mobile node, our algorithm has a main advantage: as n grows, it remains correct with smaller values of Δpredict. This feature makes the real world implementation of the neighbor detector more feasible. In fact, although there exist different approaches to predict the future locations of a node, usually predictions tend to become less accurate as Δpredictincreases.

We introduce a time-limited neighbor detector service for mobile ad hoc networks, which enables a mobile device to detect other nearby devices in the past, present and up to some bounded time interval in the future. Our motivation lies in the emergence of a new trend of mobile applications known as proximity-based mobile applications, which enable a user to communicate with other users within some defined range and for a certain amount of time. Neighbor discovery is a fundamental requirement for these applications and is not restricted to the current neighbors but can include past or future neighbors. To implement the time-limited neighbor detector service, we apply an approach based on virtual mobile nodes. A virtual mobile node is an abstraction that is akin to a mobile node that travels in the network in a predefined trajectory. In practice it can be implemented by a set of mobile nodes based on a replicated state machine approach. In this paper, we assume that each node can accurately predict its own locations up to some bounded time interval in the future. Thus, we present a time-limited neighbor detector algorithm that uses a virtual mobile node that continuously travels in the network, collects the predicted locations of all nodes, performs the neighborhood matching between nodes and sends the list of neighbors to each node. We show that our algorithm correctly implements the time-limited neighbor detector service under a set of conditions.

This paper introduces EAGLEMACAW, a replication-based protocol for streaming media content to consumers in P2P systems, in an efficient and reliable manner. To guarantee efficiency and reliability, EAGLEMACAW adaptively replicates media content in different peers while streaming to consumers. In addition, two distinct trees, named EAGLE TREE and MACAW TREE, are built on top of an overlay network to route media content to consumers. The EAGLE TREE is built based on efficiency metrics while the MACAW TREE is constructed based on reliability metrics. The EAGLE TREE is used as a main routing path for streaming media content to consumers with the goal of using the network resources in an efficient manner. As soon as a partitioning occurs in the EAGLE TREE, the MACAW TREE is used as a temporary path to continue streaming to consumers. Both these trees are built in a distributed manner, which ensures the scalability of the approach. Our performance evaluation shows that in an unreliable environment, over 90% of consumers on average are served via the EAGLE TREE. It also shows that when partitioning occurs in the EAGLE TREE, by switching to the MACAW TREE, consumers are capable of consuming around 50% more media content on average than they would in absence of the MACAW TREE.

Message ordering is one of the cornerstones of reliable distributed systems. However, some ordering guarantees, such as atomic order, are expensive to implement in terms of message delays. This paper presents Optimistic Atomic Multicast, a protocol that combines reduced latency and increased throughput. Messages can be delivered optimistically in a single communication step and conservatively in three communication steps. Differently from previous optimistic group communication protocols, Optimistic Atomic Multicast does not rely on spontaneous message ordering for fast delivery. In addition to presenting Optimistic Atomic Multicast, we provide detailed performance results comparing it to other ordering protocols in both local-area and wide-area networks.

In this paper, we consider the problem of improving the detection of a device by another device in mobile ad hoc networks, given a maximum amount of time that they remain in proximity of each other. Our motivation lies in the emergence of a new trend of mobile applications known as proximity-based mobile applications which enable a user to communicate with other users in some defined range and for a certain amount of time. The highly dynamic nature of these applications makes neighbor detection time-constrained, i.e., even if a device remains in proximity for a limited amount of time, it should be detected with a high probability as a neighbor. To address this problem, we perform a realistic simulation-based study in mobile ad hoc networks and we consider three typical urban environments where proximity-based mobile applications are used, namely indoor with hard partitions, indoor with soft partitions and outdoor urban areas. In our study, a node periodically broadcasts a message in order be detected as a neighbor. Thus, we study the effect of parameters that we believe could influence the detection probability, i.e., the transmission power and the time interval between two consecutive broadcasts. More precisely, for each environment, we determine when a change in the value of each of these parameters could lead to an improvement of the neighbor detection and when it hurts. Our experiments show that there exists no unique combination of values of these parameters that maximizes the detection probability in all environments. Accordingly, for each environment, we present the combination that maximizes the detection probability in that environment.

Multimedia consumption over the Internet is emerging as one of the largest sink of network resources, making scalable and reliable streaming increasingly challenging. To address this challenge, we propose ReStream, an adaptive replication algorithm that relies on replication to achieve reliable and scalable streaming in resource-constrained environments. Our algorithm dynamically adapts replica placement to maximize the number of consumers under latency and bandwidth constraints, while minimizing the number of replicas. In addition, ReStream supports partitioning, i.e., replicas can be located anywhere in the network and do not necessarily form a connected graph. This allows ReStream to yield the same performance in consumption models where consumers tend to be geographically co-located, as well as in consumption models where consumers placement is totally random.

Location-based multicast addressing (LMA) yields an important building block for context-aware applications in mobile ad hoc networks (MANETs). In LMA, messages are routed based on their content as well as on the location of the sending and the receiving nodes. The same dynamism that motivates locations as part of the addressing mechanism for multicast applications in MANETs, makes such a multicast challenging to implement both efficiently and reliably across application scenarios. Different implementation strategies have been proposed in literature for abstractions similar to LMA, motivated and validated by specific applications. The goal of this paper is to devise specific implementation strategies for LMA and compare these strategies in the context of several application scenarios, in order to aid in the selection of a scheme for a given application. To that end, we first detail three algorithms for implementing LMA. The first, message-centric, strategy uses geographically scoped gossiping to propagate messages. The second, query-centric, strategy propagates queries of receivers to subsequently route messages. The third, hybrid, strategy strives for the best of both worlds through a restricted multicasting of both messages and queries. We compare these algorithms both analytically and empirically. We pinpoint differences and break-even points among the approaches based on communication patterns, contrasting our findings with common expectations and our analysis. Our evaluations show that the hybrid approach invariably outperforms at least one of the other approaches, making it a safe choice for settings with varying or unknown communication patterns.

We introduce spot cast, a new communication abstraction specifically aimed at the development of mobile proximity-based applications running in mobile ad hoc networks (MANETs). Our motivation lies in the fact that traditional communication abstractions, typically broadcast primitives with strong consistency guarantees, do not adequately capture the intrinsic here-and-now nature of such applications. Rather, developers need a communication abstraction offering the notion of proximity-based diffusion and some level of message durability, which is precisely what spot cast provides. We illustrate how spot cast can be used to implement mobile applications and we shortly discuss the correctness and the implement ability of the spot cast abstraction in MANETs.

In this paper, we propose Scale Stream, a new algorithm to support scalable multimedia streaming in peer-to-peer networks, under strict resources constraints. With the growth of multimedia consumption over the Internet, achieving scalability in a resource-constrained environment is indeed becoming a critical requirement. Intuitively, our approach consists in dynamically replicating the multimedia content being streamed, based on bandwidth and memory constraints. Our replication management strategy maximizes the number of consumers being served concurrently, while minimizing memory usage, under strict bandwidth constraints.

Broadcast is an important building block in ad hoc networks. Its challenge is to deliver a message to all nodes in the network for a reasonable cost in terms of message load and delay. Several context-aware broadcasting protocols have been proposed in order to meet this challenge, using location or proximity information in order to fine-tune retransmission decisions. However, existing protocols often target one specific setting and can reveal to be sub-optimal when settings change. Typically, optimal parameters for dense networks will differ from optimal parameters for sparse networks. To address this issue, we propose PLAN-B an adaptive proximity-based broadcast protocol that offers the possibility to define policies in order to adapt its parameters for different network settings at runtime. Our performance evaluations show that PLAN-B outperforms existing static and adaptive protocols in terms of message load in changing and unknown densities up to a factor of 2.

In this paper, we introduce six-shot multicast (6Shot), a location-aware multicast algorithm devised for mobile ad hoc networks. Multicast is a one-to-many communication scheme and has largely been studied in such networks. Indeed, this communication primitive can be used as a building block for popular services, such as data streaming or group communication. The particularity of 6Shot is the location- aware routing scheme of its implementation, which offers improved efficiency in terms of message overhead compared to existing algorithms, for a reasonable cost in terms of reliability.

Recently there has been an effort to build scalable and reliable application-level multicast solutions that combine the resilience of pure gossip-based with the efficiency of tree-based schemes. However, such solutions assume that participants have unlimited resources, for instance, that they can send an unbounded number of messages to mask network omissions. Such scenario is not realistic, specially for streaming protocols, where messages can be transmitted at a very high rate and have a small temporal validity. In this paper, we propose RASM, a scalable distributed protocol for application-level multicast. Our protocol is based on the combination of gossip-based and tree-based multicast schemes. Unlike previous approaches, which strive to combine gossip-based and tree-based schemes, our solution takes into consideration the reliability of components: nodes and communication links can fail, unexpectedly, ceasing their operation and dropping messages, respectively. Experimental results show that our scheme offers better reliability than previous solutions with low overhead.

In this paper, we propose a resource-aware solution to achieving reliable and scalable stream diffusion in a probabilistic model, i.e. where communication links and processes are subject to message losses and crashes, respectively. Our solution is resource-aware in the sense that it limits the memory consumption, by strictly scoping the knowledge each process has about the system, and the bandwidth available to each process, by assigning a fixed quota of messages to each process. We describe our approach as gambling in the sense that it consists in accepting to give up on a few processes sometimes, in the hope of better serving all processes most of the time. That is, our solution deliberately takes the risk not to reach some processes in some executions, in order to reach every process in most executions. The underlying stream diffusion algorithm is based on a tree-construction technique that dynamically distributes the load of forwarding stream packets among processes, based on their respective available bandwidths. Simulations show that this approach pays off when compared to traditional gossiping, when the latter faces identical bandwidth constraints.

This paper presents QuoCast, a resource-aware protocol for reliable stream diffusion in unreliable environments, where processes may crash and communication links may lose messages. QuoCast is resource-aware in the sense that it takes into account memory, CPU, and bandwidth constraints. Memory constraints are captured by the limited knowledge each process has of its neighborhood. CPU and bandwidth constraints are captured by a fixed quota on the number of messages that a process can use for streaming. Both incoming and outgoing traffic are accounted for. QuoCast maximizes the probability that each streamed packet reaches all consumers while respecting their incoming and outgoing quotas. The algorithm is based on a tree-construction technique that dynamically distributes the forwarding load among processes and links, based on their reliabilities and on their available quotas. The evaluation results show that the adaptiveness of QuoCast to several contraints provides better reliability when compared to other adaptive approaches.

We propose Streamline, a two-layered architecture designed for media streaming in overlay networks. The first layer is a generic, customizable and lightweight protocol which is able to construct and maintain different types of meshes, exhibiting different properties. We discuss two types of overlay networks and explain how the first layer protocol builds these networks in a distributed manner. The second layer is responsible for data propagation to the nodes in the mesh by constructing an optimized diffusion tree. In order to cover the vulnerabilities of the diffusion tree, we propose a masking mechanism which enables the nodes to instantly switch to alternative data paths when necessary. Our simulations reveal that, the structure and properties of the underlying mesh are key to the performance of the system and Streamline can tolerate high node churn without degrading delivery rate.

We show that scale-free topologies have a positive impact on the performance of gossiping algorithms in peer-to-peer overlay networks. This result is important in the context of ad hoc networks, where each node participates in controlling the network topology. Our study shows that, when combined with such topologies, typical gossiping algorithms tend to require fewer messages and experience smaller latency than when combined with other topologies, such as rings or grids. This suggests that the topology control scheme should aim at producing an overlay network that exhibits scale-free characteristics.

In this paper, we propose a resource-aware solution to achieving reliable and scalable stream diffusion in a probabilistic model, i.e., where communication links and processes are subject to message losses and crashes, respectively. Our solution is resource-aware in the sense that it limits the memory consumption, by strictly scoping the knowledge each process has about the system, and the bandwidth available to each process, by assigning a fixed quota of messages to each process. We describe our approach as gambling in the sense that it consists in accepting to give up on a few processes sometimes, in the hope to better serve all processes most of the time. That is, our solution deliberately takes the risk not to reach some processes in some executions, in order to reach every process in most executions. The underlying stream diffusion algorithm is based on a tree-construction technique that dynamically distributes the load of forwarding stream packets among processes, based on their respective available bandwidths. Simulations show that this approach pays off when compared to traditional gossiping, when the latter faces identical bandwidth constraints.

This paper introduces Pervaho, a platform for developing and testing mobile ad hoc applications. The Pervaho platform is founded on a location-based publish/subscribe service, which allows mobile peers to interact in a flexible and anonymous manner, based on their collocation at the time of the interaction. To validate the semantics of location-based criteria-a general issue in mobile ad hoc applications-we also propose a phone motion simulator as second cornerstone of our platform. Finally, we evaluate the usability of Pervaho by developing a concrete application with and without Pervaho

This paper introduces the concept of location-based publish/subscribe (LPS), which allows mobile ad hoc applications to anonymously communicate with each other, depending on their locations. With this concept, publish/subscribe topics are typically expressed in a dynamic manner including proximity criteria, e.g., "I subscribe to all events on topic T published within some range R". We advocate that location-based publish/subscribe is a key programming paradigm for building mobile ad hoc application, and sketch our current implementation, which is based on standard APIs of the Java 2 platform, Micro Edition

In this paper, we propose a novel approach for solving the reliable broadcast problem in a probabilistic unreliable model. Our approach consists in first defining the optimality of probabilistic reliable broadcast algorithms and the adaptiveness of algorithms that aim at converging toward such optimality. Then, we propose an algorithm that precisely converges toward the optimal behavior, thanks to an adaptive strategy based on Bayesian statistical inference. We compare the performance of our algorithm with that of a typical gossip algorithm through simulation. Our results show, for example, that our adaptive algorithm quickly converges toward such exact knowledge.

DCL

The GARF, Bast, and OGS systems represent the resulting efforts of a multi-year ‘object group’ program at the Swiss Federal Institute of Technology in Lausanne. The intent of the program was to understand the extent to which one could build flexible and performance system supports to encapsulate object plurality. That is, we experimented with various ways to build libraries and services to support object groups in a distributed setting. This paper summarizes the main steps of the efforts and draws some conclusions about the successes and failures of our approaches. Copyright © 2000 John Wiley & Sons, Ltd.

The paper presents BAST, an object-oriented library of reliable distributed protocols. The authors show how BAST can be used to build fault-tolerant distributed applications, and how new protocols can be added to it. They discuss some distributed protocol design issues and the way these issues are circumvented in BAST. They briefly describe their Smalltalk and Java implementations of BAST, together with some performance results.

This paper draws several observations from our experiences in building support for object groups. These observations actually go beyond our experiences and may apply to many other developments of object based distributed systems.Our first experience aimed at building support for Smalltalk object replication using the Isis process group toolkit. It was quite easy to achieve group transparency but we were confronted with a strong mismatch between the rigidity of the process group model and the flexible nature of object interactions. Consequently, we decided to build our own object oriented protocol framework, specifically dedicated to support object groups (instead of using a process group toolkit). We built our framework in such a way that basic distributed protocols, such as failure detection and multicasts, are considered as first class entities, directly accessible to the programmers. To achieve flexible and dynamic protocol composition, we had to go beyond inheritance and objectify distributed algorithms.Our second experience consisted in building a CORBA service aimed at managing group of objects written on different languages and running on different platforms. This experience revealed a mismatch between the asynchrony of group protocols and the synchrony of standard CORBA interaction mechanisms, which limited the portability of our CORBA object group service. We restricted the impact of this mismatch by encapsulating asynchrony issues inside a specific messaging sub-service.We dissect the cost of object group transparency in our various implementations, and we point out the recurrent sources of overheads, namely message indirection, marshaling/unmarshaling and strong consistency.

The common object request broker architecture (CORBA) is becoming a standard for distributed application middleware, and there are increasing needs for enriching the basic functionalities of CORBA. While mechanisms for persistence, transactions, event channels, etc. have been designed and specified for CORBA, no standard support is provided to handle object replication. In this paper we discuss the issue of augmenting CORBA with group communication, which is considered an adequate paradigm to handle replication. We distinguish two main approaches: the integration approach and the service approach. We argue that the service approach is more appropriate to CORBA as it preserves the modularity of the architecture. We describe a proposal for a group communication service and discuss some implementation issues.