3.1 Introduction

Our scientific foundations are grounded on distributed collaborative systems supported by sophisticated data sharing mechanisms and on service oriented computing with an emphasis on orchestration and on non-functional properties. Distributed collaborative systems enable distributed group work supported by computer technologies. Designing such systems requires an expertise in Distributed Systems and in Computer-supported collaborative Work. Besides theoretical and technical aspects of distributed systems, the design of distributed collaborative systems must take into account the human factor to offer solutions suitable for users and groups.

The Coast team vision is to move away from a centralized authority based collaboration toward a decentralized collaboration. Users will have full control over their data. They can store them locally and decide with whom to share them. The Coast team investigates the issues related to the management of distributed shared data and coordination between users and groups.

Service oriented Computing  26 is an established domain on which the ECOO, Score and now the Coast teams have been contributing for a long time. It refers to the general discipline that studies the development of computer applications on the web. A service is an independent software program with a specific functional context and capabilities published as a service contract (or more traditionally an API). A service composition aggregates a set of services and coordinates their interactions. The scale, the autonomy of services, the heterogeneity and some design principles underlying Service Oriented Computing open new research questions that are at the basis of our research. They span the disciplines of distributed computing, software engineering and computer supported collaborative work (CSCW). Our approach to contribute to the general vision of Service Oriented Computing is to focus on the issue of the efficient and flexible construction of reliable and secure high-level services. We aim to achieve it through the coordination/orchestration/composition of other services provided by distributed organizations or people.

3.2 Consistency Models for Distributed Collaborative Systems

Collaborative systems are distributed systems that allow users to share data. One important issue is to manage the consistency of shared data according to concurrent access. Traditional consistency criteria such as serializability, linearizability are not adequate for collaborative systems. Causality, Convergence and Intention preservation (CCI)  30 are more suitable for developing middleware for collaborative applications. We develop algorithms for ensuring CCI properties on collaborative distributed systems. Constraints on the algorithms are different according to the kind of distributed system and to the data structure. The distributed system can be centralized, decentralized or peer-to-peer. The type of data can include strings, growable arrays, ordered trees, semantic graphs and multimedia data.

3.3 Optimistic Replication

Replication of data among different nodes of a network promotes reliability, fault tolerance, and availability. When data are mutable, consistency among the different replicas must be ensured. Pessimistic replication is based on the principle of single-copy consistency while optimistic replication allows the replicas to diverge during a short time period. The consistency model for optimistic replication  28 is called eventual consistency, meaning that replicas are guaranteed to converge to the same value when the system is idle. Our research focuses on the two most promising families of optimistic replication algorithms for ensuring CCI:

• operational transformation (OT) algorithms  21
• algorithms based on commutative replicated data types (CRDT)  27.

Operational transformation algorithms are based on the application of a transformation function when a remote modification is integrated into the local document. Integration algorithms are generic, being parametrised by operational transformation functions which depend on replicated document types. The advantage of these algorithms is their genericity. These algorithms can be applied to any data type and they can merge heterogeneous data in a uniform manner. Commutative replicated data types are a new class of algorithms initiated by WooT 25, the first algorithm designed WithOut Operational Transformations. They ensure consistency of highly dynamic content on peer-to-peer networks. Unlike traditional optimistic replication algorithms, they can ensure consistency without concurrency control. CRDT algorithms rely on natively commutative operations defined on abstract data types such as lists or ordered trees. Thus, they do not require a merge algorithm or an integration procedure.

3.4 Process Orchestration and Management

Process Orchestration and Management is considered as a core discipline behind Service Management and Computing. It includes the analysis, the modelling, the execution, the monitoring and the continuous improvement of enterprise processes and is for us a central domain of study. Many efforts have been devoted to establishing standard business process models founded on well-grounded theories (e.g. Petri Nets) that meet the needs of business analysts, software engineers and software integrators. This has led to heated debate in the Business Process Management (BPM) community as the two points of view are very difficult to reconcile. On one side, business people in general require models that are easy to use and understand and that can be quickly adapted to exceptional situations. On the other side, IT people need models with an operational semantics in order to be able transform them into executable artifacts. Part of our work has been an attempt to reconcile these points of view. This resulted in the development of the Bonita BPM system. It also resulted more recently in our work on crisis management where the same people are designing, executing and monitoring the process as it executes. More generally, and at a larger scale, we have been considering the problem of processes spanning the barriers of organizations. This leads to the more general problem of service composition as a way to coordinate inter organizational construction of applications. These applications provide value, based on the composition of lower level services  18.

3.5 Service Composition

Recently, we started a study on service composition for software architects where services are coming from different providers with different plans (capacity, degree of resilience...). The objective is to support the architects to select the most accurate services (w.r.t. to their requirements, both functional and non-functional) and plans for building their software. We also compute the properties that we enforce for the composition of these services.

4.1 Crisis Management

Crisis management research investigates all the dimensions regarding the management of unexpected catastrophic events like floods, earthquakes, terrorist attacks or pandemics. All the phases of a crisis, from preparedness to recovery require collaboration between people from many organizations. This provides opportunities to study inter-organizational collaboration at a large scale and to propose and evaluate mechanisms that ensure secure and safe collaboration. The PhD thesis of Béatrice Linot supervised by François Charoy and Jérôme Dinet and defended in 2021 provided us with a deep understanding of the factors that encourage collaboration and help to maintain trustworthy collaboration between stakeholders. This work is continued by the PhD thesis of Clélie Amiot who studies the effects of human chat-bot collaboration in this kind of setting.

4.2 Collaborative Editing

Collaborative editing is a common application of optimistic replication in distributed settings. The goal of collaborative editors, irrespective of the kind of document, is to allow a group of users to update a document concurrently while ensuring that they eventually get all the same copy at the end. Our algorithm allows us to implement a collaborative editor in a peer to peer way. It avoids the need for a central server ensuring a higher level of privacy among collaborators. In this context, it requires us to consider the problem of access control of participants 15.

6.1 New software

7.1 Overcoming Identifier Overhead in Sequence CRDTs

Participants: Matthieu Nicolas, Gérald Oster, Olivier Perrin.

In order to achieve high availability, large-scale distributed systems often replicate data and minimize coordination among nodes. One approach that has gained attention in both literature and industry is the use of Conflict-free Replicated Data Types (CRDTs) to design these systems. CRDTs are new specifications of existing data types, such as sets or sequences, that maintain the same behavior as previous specifications in sequential executions, but excel in distributed settings by natively supporting concurrent updates. To accomplish this, CRDTs embed conflict resolution mechanisms into their specifications. These mechanisms typically rely on identifiers attached to elements of the data structure to resolve conflicts in a deterministic and coordination-free manner. However, these identifiers must comply with certain constraints, such as uniqueness or belonging to a dense total order, which can increase the size of the identifiers over time and cause performance issues. To address this problem, we propose a novel Sequence CRDT 10 that incorporates a renaming mechanism, allowing nodes to reassign shorter identifiers to elements in an uncoordinated manner.

We conducted an experimental evaluation to assess the performance of our proposed renaming mechanisms in various areas. Specifically, we evaluated the impact of the renaming mechanism on: (i) the size of the data structure; (ii) the integration time of the rename operation; and (iii) the integration time of insert and remove operations. For comparison, we used LogootSplit 17 (see section 6.1.1) as the baseline data structure in cases (i) and (iii). The results we obtained were very promising, as we observed a significant reduction in integration time when using the renaming mechanism, even when taking into account the time required to execute the rename operation.

7.2 Distributed Access Control using CRDTs

Participants: Claudia-Lavinia Ignat, Olivier Perrin, Pierre-Antoine Rault.

Existing access control mechanisms mainly based on a central authority feature several difficulties in the context of collaborative systems. In the case of a federation of organizations, agreeing on an authority that manages the access rights is almost impossible. Lack of a central authority raises issues of group management such as joining and leaving the group as well as rights revocation. Moreover, current access control mechanisms feature performance issues that are critical for real-time collaboration when the number of updates is high. Indeed, delays are too high for sending at each user action an access request and waiting for its answer from a trusted central authority which maintains the security policies.

We proposed a distributed access control mechanism for collaborative applications where access rights as well as data are replicated 15. We illustrated by means of examples the challenges faced by replication algorithms for obtaining consistency over both data and access control policies in distributed applications such as Google Docs and POSIX file systems. We provided a replication algorithm for access control policies based on Conflict-free Replicated Data Types (CRDTs). We also gave an overview of our proposed solution for the composition of the CRDT on access control policies with a CRDT on data.

7.3 Analysis of Social Networks as Collaboration Support

Participants: Quentin Laporte Chabasse, Gérald Oster, François Charoy.

Creating secure peer-to-peer collaborative services requires a trusted peer-to-peer network. In order to achieve this goal, we examined how to utilize the underlying social networks of inter-organizational collaboration to support such collaboration. To accomplish this, we analyzed collaborative graphs, which provide valuable insights into the behavior of groups of individuals. Exponential Random Graph Models (ERGMs) are commonly used to analyze social processes and dependencies among group members. Our approach uses a modified version of ERGMs, modeling the problem as an edge-labeling one. The main challenge is inferring the model, as the normalizing constant involved in traditional Markov Chain Monte Carlo approaches is not available in closed-form.

We proposed 9 to use the ABC Shadow algorithm  29, which can sample from posterior distributions while avoiding this limitation. The method was demonstrated using real data sets provided by the HAL platform and offers new insights into self-organized collaborations among researchers. We applied this method in a longitudinal study to identify patterns in the evaluation of collaboration over multiple years for the same teams. We also applied it to a dataset from a social study in a French primary school.

7.4 Ethereum’s Peer-to-Peer Network Monitoring

Participants: Olivier Perrin, Thibault Cholez , Jean-Philippe Eisenbarth .

The two main blockchains, Bitcoin and Ethereum, rely on peer-to-peer networks to ensure their functionalities. In the work carried out with Jean-Philippe Eisenbarth and Thibault Cholez (from the RESIST team), we first proposed an in-depth study of these networks using a supervision mechanism. In particular, we have studied a number of criteria related to reliability, such as the number of peers, their geographical distribution, their distribution over the IP network, the churn, the proportion of clients with known vulnerabilities, the existence of daily connection patterns or the ability to infer topology. It appears that both networks show good properties on all these points.

Second, based on the observation that, on the one hand, the distributed hash table (DHT) of the Ethereum P2P network is largely untapped, and that, on the other hand, the storage of blockchain data is only growing (which will pose problems in the long run), we have developed a new distributed storage architecture for Ethereum that takes advantage of the DHT. The proposed solution is backward compatible with current clients and can reduce the disk space used for long-term storage by 95% (58% of total storage) without impacting the guarantees or performance of the Ethereum blockchain.

Finally, we analyzed Ethereum peers for patterns that could reflect Sybil attacks and showed the existence of thousands of suspicious nodes grouping a large number of identifiers for the same IP address (up to 10,000/IP). Following this analysis, we designed and implemented a protection architecture against Sybil attacks. It is based on a crawler detecting suspicious nodes in real time, a smart contract structuring the information and distributing it to all peers, and finally a completely distributed revocation mechanism, each peer noticing itself the attack and cutting its connections to Sybils nodes. The implementation on an Ethereum test network has shown the efficiency of the proposed architecture 7.

7.5 Composite Service Selection based on API Call Limit

Participants: François Charoy.

7.6 Straggler Detection in Big Data Analytic Systems

Participants: Thomas Lambert.

Speculative execution can significantly improve the performance of Big Data applications by launching other copies of stragglers (slow tasks). Straggler detection plays an important role in the effectiveness of speculative execution. Not detecting real stragglers (false negatives) may slow down the whole computation. On the contrary, launching speculative copies of non-straggler tasks (false positives) can lead to a waste of resources. Most state-of-the-art methods employed to detect stragglers use the information extracted from the last received heartbeats which may be outdated when triggering detection. This, in turn, can mislead Big Data analytics systems to make inaccurate detection.

In 14 we shed light on this issue by carrying out extensive simulations to identify how heartbeat arrival, task starting times, and detection methods impact the accuracy of straggler detection in Big Data analytic systems. In particular we investigated two families of state-of-the-art detection methods: one based on progression score and the other one based on progression speed. We showed that they can both lead to rather large inaccuracy, by either overlooking real stragglers or by marking normal tasks as stragglers. We also highlighted the fact that this inaccuracy is not only due to outdated information, but also to the asynchrony of starting times and heartbeat arrivals.

7.7 Privacy-Preserving Graph Processing in Geo-Distributed Data Centers

Participants: Thomas Lambert.

Graph processing is a popular computing model for big data analytics. Very roughly, in such model, data are stored on different nodes of a graph and are updated iteratively with data transferred from the neighboring nodes. The famous Page Rank algorithm, which is very popular for ranking web pages or users, is an example of distributed analytics that can use this model. Meanwhile, emerging big data applications are often maintained in multiple geographically distributed (geo-distributed) data centers (DCs) to provide low-latency services to global users. Graph processing in geo-distributed DCs suffers from costly inter-DC data communications. Furthermore, due to increasing privacy concerns, geo-distribution imposes diverse, strict, and often asymmetric privacy regulations that constrain geo-distributed graph processing. For example, the European General Data Protection Regulation (GDPR) does not allow the movement of private data from a DC in Europe to a DC in the USA. Existing graph processing systems fail to address these two challenges.

In 16, we designed and implemented PGPregel, an end-to-end system that provides privacy-preserving graph processing in geo-distributed DCs with low latency and high utility. The design of PGPregel is based on three techniques. The first technique is differential privacy, a popular way to preserve privacy by adding noise to transferred data. The second technique is sampling, i.e. only part of the data is sent from one DC to another. The objective is to reduce both inter-DC communication and the amount of added noise in order to keep the same level of privacy (which results in a possible improvement of accuracy). The last technique is combiners, i.e. aggregated messages. Similarly to sampling, combiners reduce inter-DC communication and mitigate the impact of noise on accuracy.

We implemented our design in Apache Giraph(https://giraph.apache.org/) and evaluated it in real cloud DCs. We targeted four graph processing applications. Results show that PGPregel can preserve the privacy of graph data with low overhead and good accuracy, beating strategies based on differential privacy, no-inter-DC communication, or application-specific privacy-preserving implementation alone.

7.8 Proactive IoT Business Process Management with Data Analytics and Event Processing

Participants: Khalid Benali, Abir Ismaili-Alaoui.

IoT is becoming a hot-spot area of technological innovations and economic development promises for many industries and services. This new paradigm shift affects all enterprise architecture layers, from infrastructure to business. Business Process Management (BPM) is affected by this new technology.

To tackle the data and event explosion resulting, among others, from IoT, data analytic processes are combined with event processing techniques. These techniques examine large data sets to uncover hidden patterns, and unknown correlations between collected events, either at a very technical level (incident/anomaly detection, predictive maintenance) or at the business level (customer preferences, market trends, revenue opportunities). They provide improved operational efficiency, better customer service, and competitive advantages over rival organizations.

In 8 and 13 we proposed new approaches for augmenting business processes. By relying mainly on data analysis, machine learning algorithms, and complex event processing, we exploited the data generated by business process execution (event data, event logs) and improved these processes from different perspectives such as instance scheduling and event management in an IoT environment.

8.1 Bilateral contracts with industry

9.1 National initiatives

10.1 Promoting scientific activities

10.2 Teaching - Supervision - Juries

10.3 Popularization

11.1 Major publications

• 1 articleC.-L.Claudia-Lavinia Ignat, L.Luc André and G.Gérald Oster. Enhancing rich content wikis with real-time collaboration.Concurrency and Computation: Practice and Experience338April 2021
  HALDOI
• 2 articleC.-L.Claudia-Lavinia Ignat, Q.-V.Quang-Vinh Dang and V.Valerie Shalin. The Influence of Trust Score on Cooperative Behavior.ACM Transactions on Internet Technology194September 2019, 1-22
  HALDOI
• 3 articleQ.Quentin Laporte-Chabasse, R. S.Radu S. Stoica, M.Marianne Clausel, F.François Charoy and G.Gérald Oster. Morpho-statistical description of networks through graph modelling and Bayesian inference.IEEE Transactions on Network Science and Engineering942022, 2123 - 2138
  HALDOI
• 4 articleH.Hoai Le Nguyen and C.-L.Claudia-Lavinia Ignat. An Analysis of Merge Conflicts and Resolutions in Git-based Open Source Projects.Computer Supported Cooperative Work273-6June 2018, 741-765
  HALDOI
• 5 articleM.Matthieu Nicolas, G.Gerald Oster and O.Olivier Perrin. Efficient Renaming in Sequence CRDTs.IEEE Transactions on Parallel and Distributed Systems3312December 2022, 3870-3885
  HALDOI
• 6 inproceedingsG.Guillaume Rosinosky, S.Samir Youcef and F.Francois Charoy. A Genetic Algorithm for Cost-Aware Business Processes Execution in the Cloud.Lecture Notes in Computer ScienceICSOC 2018 - The 16th International Conference on Service-Oriented ComputingICSOC 2018: Service-Oriented Computing11236Hangzhou, ChinaSpringerNovember 2018, 14
  HAL

11.2 Publications of the year

11.3 Cited publications

• 17 inproceedingsL.Luc André, S.Stéphane Martin, G.Gérald Oster and C.-L.Claudia-Lavinia Ignat. Supporting Adaptable Granularity of Changes for Massive-scale Collaborative Editing.CollaborateCom - 9th IEEE International Conference on Collaborative Computing: Networking, Applications and Worksharing - 2013Austin, United StatesOctober 2013, URL: http://hal.inria.fr/hal-00903813
  back to text
• 18 articleS.Sami Bhiri, O.Olivier Perrin, W.Walid Gaaloul and C.Claude Godart. An Object-Oriented Metamodel For Inter-Enterprises Cooperative Processes Based on Web Services.Journal of Integrated Design and Process Science82004, 37--55
  HALback to text
• 19 incollectionF.Fabio Casati. Promises and Failures of Research in Dynamic Service Composition.Seminal Contributions to Information Systems EngineeringSpringer Berlin Heidelberg2013, 235-239URL: http://dx.doi.org/10.1007/978-3-642-36926-1_18
  back to text
• 20 inproceedingsQ. V.Quang Vinh Dang and C.-L.Claudia-Lavinia Ignat. Computational Trust Model for Repeated TrustGames. Proceedings of the 15th IEEE International Conference on Trust, Security and Privacy in Computing and Communications (TrustCom 2016) Tianjin, ChinaAugust 2016, URL: https://hal.inria.fr/hal-01351250
  back to text
• 21 inproceedingsC. A.Clarence A. Ellis and S. J.Simon J. Gibbs. Concurrency Control in Groupware Systems.Proceedings of the ACM SIGMOD Conference on the Management of Data - SIGMOD 89Portland, Oregon, USAMay 1989, 399--407URL: http://doi.acm.org/10.1145/67544.66963
  back to textback to text
• 22 articleC.-L.Claudia-Lavinia Ignat, Q.-V.Quang-Vinh Dang and V.Valerie Shalin. The Influence of Trust Score on Cooperative Behavior.ACM Transactions on Internet Technology194September 2019, 1-22
  HALDOIback to text
• 23 inproceedingsH.-L.Hoang-Long Nguyen, J.-P.Jean-Philippe Eisenbarth, C.-L.Claudia-Lavinia Ignat and O.Olivier Perrin. Blockchain-Based Auditing of Transparent Log Servers.The 32nd Annual IFIP WG 11.3 Conference on Data and Applications Security and Privacy (DBSec 2018)Proceeding of Data and Applications Security and Privacy XXXII - 32nd Annual IFIP WG 11.3 ConferenceBergamo, ItalyJuly 2018, 21-37URL: https://hal.archives-ouvertes.fr/hal-01917636
  back to text
• 24 inproceedingsH.-L.Hoang-Long Nguyen, C.-L.Claudia-Lavinia Ignat and O.Olivier Perrin. Trusternity: Auditing Transparent Log Server with Blockchain.Companion of the The Web Conference 2018Lyon, FranceApril 2018, 79-80URL: https://hal.inria.fr/hal-01883589
  DOIback to text
• 25 inproceedingsG.Gérald Oster, P.Pascal Urso, P.Pascal Molli and A.Abdessamad Imine. Data Consistency for P2P Collaborative Editing.ACM Conference on Computer-Supported Cooperative Work - CSCW 2006Banff, Alberta, CanadaACM Press11 2006, 259 - 268URL: http://hal.inria.fr/inria-00108523/en/
  back to text
• 26 articleM. P.Michael P. Papazoglou, P.Paolo Traverso, S.Schahram Dustdar and F.Frank Leymann. Service-Oriented Computing: State of the Art and Research Challenges.Computer402007, 38-45
  back to text
• 27 inproceedingsN.Nuno Preguiça, J. M.Joan Manuel Marquès, M.Marc Shapiro and M.Mihai Letia. A commutative replicated data type for cooperative editing.29th IEEE International Conference on Distributed Computing Systems (ICDCS 2009)Montreal, Québec CanadaIEEE Computer Society2009, 395-403
  HALDOIback to text
• 28 articleY.Yasushi Saito and M.Marc Shapiro. Optimistic Replication.Computing Surveys371March 2005, 42--81URL: http://doi.acm.org/10.1145/1057977.1057980
  back to text
• 29 articleR. S.Radu S Stoica, A.Anne Philippe, P.Pablo Gregori and J.Jorge Mateu. ABC Shadow algorithm: a tool for statistical analysis of spatial patterns.Statistics and computing2752017, 1225--1238
  back to text
• 30 articleC.Chengzheng Sun, X.Xiaohua Jia, Y.Yanchun Zhang, Y.Yun Yang and D.David Chen. Achieving Convergence, Causality Preservation, and Intention Preservation in Real-Time Cooperative Editing Systems.ACM Transactions on Computer-Human Interaction51March 1998, 63--108URL: http://doi.acm.org/10.1145/274444.274447
  back to text