IaaS on Grid'5000 Workshop - 2012

From Grid5000
Revision as of 13:37, 23 April 2012 by Dmargery (talk | contribs) (April 23rd)

(diff) ← Older revision | Latest revision (diff) | Newer revision → (diff)
Jump to: navigation, search

To be held in Rennes, April 23rd 2012.

Call for participation

Dear all,

It is my pleasure to invite Grid'5000 users to a half day workshop in
Rennes on April 23rd, 2012, followed by an optional 2 days coding camp
for all Grid'5000 users working on the deployment of open source Iaas
stacks (OpenNebula, Nimbus, OpenStack, ..).

During the coding camp on the following days, participants will be able to
start collaborating on consolidating their tools. It is expected that people
participating in the coding camp are allowed to contribute to a common
code-base in a Cecill-B license (the one available to the Grid'5000 technical
team to share code with users).

The dates of this event are April 23rd - April 25th, and the event 
would be held in Rennes. Details will follow on [2]

The workshop part will have people present what they their scripts 
achieve and how they work, and be concluded by identifying collaboration

The coding camp on the following days would have people start collaborating. 

Interested participants should register with David Margery by mail, with a
talk title and a brief description on their current work to deploy IaaS 
stacks on Grid'5000. This description should include the stacks they
deploy, the number of physical nodes aggregated in the deployment and 
typical deployment time. We expect to be able to host all proposers, 
and will draft a schedule on April 6th 2012. Should we be overwhelmed 
by registrations, we will discuss options with proposers (extend the 
workshop to Tuesday morning - organize a selection)


David Margery
[1] http://www.cecill.info/
[2] https://www.grid5000.fr/mediawiki/index.php/IaaS_on_Grid%275000_Workshop_-_2012


April 23rd

Meeting to be held in Métivier Room at Inria - Rennes

  • 13h30 -14h : welcome
  • 14h-15h30: first session. please aim at 20mns presentations
    • 14h00: Deploying OpenNebula on Grid'5000 to extend BonFIRE (Maxence Dunnewind)
    • 14h30: Multi-site OpenNebula deployments. (Alexandra Carpen-Amarie)
    • 15h00: Deploying OpenNebula on Grid'5000 and Experimenting with MapReduce workloads on top of OpenNebula (Jonathan Rouzaud-Cornabas)
  • 15:30 coffee break
    • 16:00-17h45 : second session and wrap-up
    • 16:00: Deploying an OpenStack cloud on Grid'5000 (Sébastien Badia)
    • 16:30: Deploying a gLite infrastructure: workflows to the rescue (Thomas Buchert)
    • 17:00: Identifying possible collaboration scenarios (David Margery)

April 24th - April 25th

9h - 18h : Coding camp


Please register to David Margery as room planned has limited capacity. I will upload candidates to this page as I receive mails.

Deploying a gLite infrastructure: workflows to the rescue

gLite (http://glite.cern.ch) is the grid middleware used on the EGI production grid.
The deployment of a gLite infrastructure shares many of the issues encountered when deploying IaaS clouds.
As a result of our work on automatizing the deployment of gLite, we are able to successfully deploy its infrastructure
inside Grid'5000 using multiple sites and clusters.
The deployment time varies: from 20 minutes for a basic deployment (10 nodes), up to 3 hours (597 nodes, 6 sites, 10 clusters).
In this presentation, we will describe our current work on reimplementing our Ruby script using a workflow engine.
It will allow for improved provisioning of the deployment, simplified searching for bottlenecks, measuring real-time consumption of resources,
improved failure policy and maintainability of our solution.

Deploying an OpenStack cloud on Grid'5000

OpenStack [http://openstack.org/] is an IaaS cloud computing project by Rackspace Cloud and NASA.
In their current state, the scripts scripts allow the user to deploy the "Diablo" version on an ubuntu 11.10 image.
The deployment is done inside a kavlan-local VLAN, and we deploy the nova, glance and volumes services (and swift is planned).
Scaling has not been evaluated yet, but a run with a cloud controller and 99 computes nodes was achieved.
Time to deploy this infrastructure was approximately 8 min (not including Ubuntu deployment).

Multi-site OpenNebula deployments

 OpenNebula (http://opennebula.org/start) is a fully open-source IaaS Cloud implementation designed to provide a 
 scalable and standardized virtualization framework for managing data centers.
 The talk will introduce a set of tools designed to automatically deploy and configure OpenNebula on Grid'5000. 
 The scripts rely on Ruby and Chef to install the latest release of OpenNebula (3.2.1) from the Debian Sid 
 packages on a standard Debian environment. The deployment supports multi-site reservations, creating a single 
 Cloud with one frontend and a large resource pool that can accommodate hundreds of VMs running simultaneously 
 (up to 350 on 80 nodes belonging to 3 sites). This presentation will show how we addressed the challenges 
 related to managing the network configuration for VMs deployed on different Grid'500 sites. We will also 
 introduce a set of wrapper client scripts that hide the configuration details of VMs and automatically create 
 VM templates to allow for "one-click" large VM deployments.

Deploying OpenNebula on Grid'5000 and Experimenting with MapReduce workloads on top of OpenNebula

  • Author: Jonathan Rouzaud-Cornabas (AVALON - Labex MILyon / INRIA / LIP) User Report
Isolation between users or applications within distributed architectures is increasing important due
to multi-tenancy feature of infrastructures.
My work focuses on isolation for security and performances on large scale hierarchical distributed
architectures like Clouds and HPC.
In this presentation, I will present a ruby/chef script to automate the deployment of OpenNebula.
Furthermore, deploying OpenNebula is only the first step of the experimentation.
The purpose was to test the trade-off between performances, isolation and consolidation.
Thus, I will also present how to deploy and configure MapReduce cluster on top of OpenNebula.
I will conclude by presenting my experimentation results.
The first experiment was to test different OpenNebula configuration: which database (SQLite/MySQL), 
file systems (shared / non shared), scheduler (default / Haizea / a custom Haizea), network
(openvswitch, ebtables, etc.).
The second experiment was on MapReduce workload: deployment and configuration, performances,
isolation, consolidation.

Deploying OpenNebula on Grid'5000 to extend BonFIRE

  • Author: Maxence Dunnewind
The [http://www.bonfire-project.eu BonFIRE project] is running a permanent federated testbed to which Inria 
contributes by running a OpenNebula testbed. The BonFIRE Project supports experimentation and testing of 
innovative scenarios from the Internet of Services research community, specifically focused on the convergence 
of services and networks. BonFIRE operates a Cloud facility based on an Infrastructure as a Service delivery 
model with guidelines, policies and best practices for experimentation.
The testbed located at Inria can be dynamically extended to the Grid'5000 resources in Rennes, by dynamically 
deploying new worker nodes using Kadeploy and [[KaVLAN]]. 
Some tweaks to OpenNebula where used to enable users of these resources to quickly deploy a large number of VM.

Other participants

  • Yacine Kessaci (Monday only)
  • David Margery
  • Trong-Tuan Vu (Monday only)
  • Jean-Marc Lambert (Monday only)
  • Denis Shestakov (Monday only, no need for a badge)
  • Ines AYADI (Monday only)