transparent data sharing
At the Eurosys2009 conference, Kim-Thomas Moeller from Universitaet Duesseldorf presented XtreemOS's object sharing service in a work-in-progress talk and during the poster session (pdf). The object sharing service addresses the problem of managing shared states in distributed and parallel applications. It aims at simplifying the development of grid applications by abstracting from access location, replica management, data consistency, and fault tolerance.
EuroSys2009 poster session - Courtesy of Wanja Hofer
XtreemOS was also represented by Guillaume Pierre from Vrije Universiteit Amsterdam who presented the "Highly Available and Scalable Grid Services" (pdf) at WDDDM and Christine Morin from INRIA who gave an invited talk at HPCVirt ("Virtualization in XtreemOS Distributed Operating System: Requirements and Opportunities").
Eurosys2010 will be located in Paris, with XtreemOS's project leader Christine Morin as the conference's general chair.
Christine Morin and Gilles Muller will be EuroSys2010 general chair and PC chair respectively
- Courtesy of Wanja Hofer
Florian Mueller (UDUS) presented the paper "Transactional Data Sharing in Grids" related to the XtreemOS Object Sharing Service (OSS), at the IASTED International Conference on Parallel and Distributed Computing and Systems in Orlando, USA.
"Transactional Data Sharing in Grids" - M.-F. Mueller, K.-T. Moeller, M. Sonnenfroh, M. Schoettner (UDUS)
Paper abstract: "The EU-funded XtreemOS project implements a Linux-based grid operating system (OS), exploiting resources of virtual organizations through the standard POSIX interface.
The Object Sharing Service (OSS) of XtreemOS addresses the challenges of transparent data sharing for distributed applications running in grids. We focus on the problem of handling consistency of replicated data in wide area networks in the presence of failures. The software architecture we propose interweaves concepts from transactional memory and peer-to-peer systems. Speculative transactions relieve programmers from complicated lock management.
Super-peer-based overlay networks improve scalability and distributed hash tables speed up data search. OSS replicates objects to improve reliability and performance. In case of severe faults, the XtreemOS grid checkpointing service will support OSS. In this paper we describe the software architecture of OSS, design decisions, and evaluation results of preliminary experiments with a multi-user 3D virtual world. "