Section: New Results

Dependable Cloud Computing

Participants : Roberto-Gioacchino Cascella, Stefania Costache, Florian Dudouet, Eugen Feller, Filippo Gaudenzi, Yvon Jégou, Ancuta Iordache, David Margery, Christine Morin, Anne-Cécile Orgerie, Guillaume Pierre, Nikos Parlavantzas, Yann Radenac, Matthieu Simonin, Cédric Tedeschi.

Multi-data Center and Multi-cloud

Deployment of distributed applications in a multi-provider environment

Participants : Roberto-Gioacchino Cascella, Stefania Costache, Florian Dudouet, Piyush Harsh, Filippo Gaudenzi, Yvon Jégou, Christine Morin.

The move of users and organizations to Cloud computing will become possible when they will be able to exploit their own applications, applications and services provided by cloud providers as well as applications from third party providers in a trustful way on different cloud infrastructures. In the framework of the Contrail European project [39]   [50] , we have designed and implemented the Virtual Execution Platform (VEP) service in charge of managing the whole life cycle of OVF distributed applications under Service Level Agreement rules on different infrastructure providers  [51] . In 2013, we designed the CIMI inspired REST-API for VEP 2.0 with support for Constrained Execution Environment (CEE), advance reservation and scheduling service, and support for SLAs  [55] , [54] [56] . We integrated support for delegated certificates and developed test scripts to integrate the Virtual Infrastructure Network (VIN) service. VEP 1.1 was slightly modified to integrate the usage control (Policy Enforcement Point (PEP)) solution developed by CNR. The CEE management interface was developed during 2013 and is available through the graphical API as well as through the RESTful API.

Towards a distributed cloud inside the backbone

Participants : Anne-Cécile Orgerie, Cédric Tedeschi.

The DISCOVERY proposal currently in phase of construction and lead by Adrien Lèbre from ASCOLA team, and currently on leave at Inria aims at designing a distributed cloud, leveraging the resources we can find in the network's backbone. (The DISCOVERY website: http://beyondtheclouds.github.io )

In this context, and in collaboration with ASCOLA and ASAP teams, we started the design of an overlay network whose purpose is to be able, with a limited cost, to locate geographically-close nodes from any point of the network. The basis for this overlay is described as part of a recent research report [44] .

Multi-cloud application deployment in ConPaaS

Participants : Guillaume Pierre, Yann Radenac.

We extended ConPaaS to support application deployment over multiple clouds. There are two main reasons for this: first, it is a necessary mechanism to allow application migration from one cloud to another, without any service interruption. Second, for some applications it may be useful to execute over multiple clouds on a permanent basis, for reliability reasons for example. The main challenges to address were ensuring full network connectivity between resources acquired in multiple clouds. We addressed these issues by integrating the IPOP virtual network in ConPaaS. Second, we designed protocols to ensure application and data migration without any service interruption during the migration.

Scalability of Snooze Self-healing Cloud Management System

Participants : Eugen Feller, Yvon Jégou, David Margery, Christine Morin, Anne-Cécile Orgerie, Matthieu Simonin.

We evaluated the scalability and resilience of Snooze IaaS management system [26] . Unlike existing systems, for scalability, ease of configuration, and high availability, Snooze is based on a self-organizing and self-healing hierarchical architecture of system services [36] , [27] , [27] . In Snooze hierarchy, each compute server is managed by a local controller that interacts with one of the group managers to which it is dynamically assigned and the set of group managers is coordinated by a group leader elected among them. We performed an extensive scalability study of Snooze across over 500 servers of the Grid’5000 experimentation testbed. We evaluated the Snooze self-organizing and self-healing hierarchy with thousands of system services. The results show that the resource consumption of the Snooze system services is bounded both during the hierarchy construction and system operation. We also show that Snooze prototype implementation is robust enough to manage thousands of servers and hundreds of VMs. Moreover, its autonomic behavior allows to achieve high availability in the presence of a large number of simultaneous system services failures. Indeed, as long as at least two group managers remain operational the system remains alive. We also demonstrated the application deployment scalability across hundreds VMs on the example of a Hadoop MapReduce application. We participated in the Scale Challenge organized in the framework of the ACM/IEEE CC-Grid 2013 conference [26] and won the second prize.

Application Performance Modeling in Heterogeneous Cloud Environments

Participants : Ancuta Iordache, Guillaume Pierre.

Heterogeneous cloud platforms offer many possibilities for applications for make fine-grained choice over the types of resources they execute on. This opens for example opportunities for fine-grain control of the tradeoff between expensive resources likely to deliver high levels of performance, and slower resources likely to cost less. We designed a methodology for automatically exploring this performance vs. cost tradeoff when an arbitrary application is submitted to the platform. Thereafter, the system can automatically select the set of resources which is likely to implement the tradeoff specified by the user. A publication on this topic is currently in preparation.

Flexible SLA & SLO Management

Participants : Stefania Costache, Christine Morin, Nikos Parlavantzas.

Merkat is a market-based, SLO-driven, PaaS system for private clouds. Merkat dynamically shares resources between competing applications to ensure a fair resource utilization in terms of application priority and actual resource needs. Resources are allocated through a proportional-share auction while autonomous controllers apply elasticity rules to scale application demand according to resource availability and user priority. Merkat provides users the flexibility to adapt controllers to their application types, and it can support diverse application types and performance goals. Merkat is implemented in Python and uses OpenNebula for virtual machine operations.

We evaluated Merkat in simulation and we analyzed the behavior of the system for multiple user types [23] . Furthermore, we deployed Merkat on Grid'5000 and EDF's tested and tested it with applications representative to EDF [22] . Results showed that: (i) the system provides flexible support for different application types (static and malleable) and different SLOs (deadline and performance); (ii) the system provides good user satisfaction achieving acceptable performance degradation, compared to existing centralized solutions. Furthermore, we extended Merkat to manage different clusters and run MPI applications on them. We also submitted a survey on evolution of resource management systems for shared virtualized computing infrastructures to an international journal. This work was carried out in the framework of Stefania Costache's PhD thesis [11] .