Snooze
autonomic virtual machines system management
Matthieu Simonin
October 15, 2013
The system
Hierarchy
-
Local Controller
Compute nodes (Host virtual machines) -
Group Manager
Manage Local Controller
VMs placement & life-cycle
Energy management -
Group Leader
Oversees Group Managers
The system
Key Features
- Self-Configuration
- Self-Healing
- Virtual machines consolidation
- Overload/underload detection
The system
Self-configuration
Step 1: Leader Election
- at system startup
- after leader failure
Step 2: Group managers join procedure
Step 2: Group managers join procedure
Step 2: Group managers join procedure
Step 2': Local Controllers join procedure
Step 2': Local Controllers join procedure
Step 2': Local Controllers join procedure
Step 2': Local Controllers join procedure
Step 2': Local Controllers join procedure
The system
Self-healing
Group Manager Failure
Group Manager Failure
Group Manager Failure
Group Manager Failure
Group Leader Failure
Group Leader Failure
Group Leader Failure
... and the promotion of the Group Manager is seen as a failure by its previously assigned Local Controller
The system
Automatic consolidation
Before consolidation
- Group Leader
- Group Manager
- Local Controller
- Virtual machine
After consolidation
From your laptop...
... to large scale
Largest deployment :
1 000 GMs
10 000 LCs
1 000 GMs
10 000 LCs
From your laptop...
... to large scale
Recovery of catastrophic failures
Developper's corner
-
Add your own policies for
- VM scheduling
- Consolidation
- Underload and overload detection
-
Control the system with the API
- get nodes monitoring information
- control vm life cycle
- trigger energy management
Source code is avalaibable in github
