Title : Infrastructure and Experimentation

Code : 4

Responsible : GDD

Activities : Performance evaluation of provided solutions

Start Date : 2013-12-01

End Date : 2017-11-30

Objectives : We will perform both theoretical and practical evaluation of the solutions proposed in the other tasks. This task will be structured around use cases where partners developed previous ex- pertise such as distributed collaborative systems (GDD with ANR Streams, RNTL XWiki-concerto), social web (Asap with ERC Gossple and soft- ware Whatsup) and Smart Building (BD with CNRS SoCQ4Home project). These well known dis- tributed systems will be revisited to fit federation of plug constraints using results of other tasks. Next, Theoretical evaluation in terms of time and space complexities will be done. Finally, we will experi- ment these solution on the cluster of Raspberry Pis of LINA (see figure). The cluster is currently com- posed by 48 Raspberry Pis and are ready to deploy SocioPlug applications.


Scientific Challenges : Three main challenges has been identified:

  • In the context of social web, we envision how software such Whatsup can take advantages of emergent localities of federation of plugs. We aim to provide a service like Goggle Flu in completely decentralized setting. By aggregating data on social-network feeds, or made available by users under the guarantee that it will remain private, we will, for example, design protocols that can track the evolution of a disease in real time.
  • In the context of distributed collaborative systems, Google doc is representative of infor- mation asymmetry and personal cloud services. In SocioPlug, we aim to rewrite an al- ternative to google doc on a federation of plugs. Same techniques can be used to rewrite equivalent of google drive, google plus, ...

Google doc is challenging for Task 1. It requires to write to have a quasi-CRDT for managing sequence that ensure eventual consistency. Such CRDT should offer bounded identifiers and resist to quasi-causality errors using original recovery protocols. This is expected result of task 1. Google doc is also challenging for componentization. It requires to efficiently discover connected users looking at this document and broadcast in real time operations to these users. Low churn of plug-federation and fixed topology can be used to optimize greatly state of art algorithms.

Next, Google doc is challenging for Task 2. According to visibility levels decided by users and meta-data available on documents, it should be possible to query efficiently the whole federation about all documents concerning a particular topic. The results set should include personal document, document shared by other but visible by me and public documents.

Finally, Google doc is challenging for Task 3. Documents should be shared with others through contracts fixing usages authorized on the particular document. Usage rules can include no-indexing, no distribution or deletion rules. The system should be in change to control if usage control is ensured by each member of federation. Finally, monitoring proto- col will try to detect any malicious activity on plugs by analyzing streams of events on each plug. We plan to write such application and deploy it on cluster of Raspberry Pis available in Nantes (see figure). The cluster is currently composed by 48 Raspberry Pis and are ready to deploy SocioPlug applications.

  • In the context of smart buildings, quarters and towns, we plan to evaluate how federation of plugs and community queries can be used to provide additional services. In this setup, nu- merous and various sensors feed the system with data streams. These streams can represent a large amount of data which can neither be transferred nor centrally treated for perfor- mance and privacy reasons. To solve the performance issue, the classical approaches are based on hierarchical network organization to aggregate data and bring them to a central final data center. We propose to investigate advantages of a queries driven organization for performance, privacy and exploitation of these data. These streamed data have to be released using usage control policies of Task 3. To perform information discovery and mon- itoring, communities organization with respect to continuous queries on streams issued by users will be handled by Task 2.

Deliverables :

  description Dec. 2013 + months
D41 Application scenarii and Design of infrastructure. 9
D42 Theoretical evaluation results of proposed solutions 39
D43 Demonstrator for extracting practical experimentation results of proposed solutions 48

Sub-tasks :

Task41 Application scenarii and Design of infrastructure
Task42 Theoretical evaluation results of proposed solutions
Task43 Practical experimentation results of proposed solutions

Participants :