Fabien Hermenier, Sophie Demassey, and Xavier Lorca. In Proceedings of the 17th international conference on Principles and practice of constraint programming (CP'11). Springer-Verlag, Berlin, Heidelberg, pages 27-41.

1. BIN REPACKING SCHEDULING
IN
VIRTUALIZED DATACENTERS
Fabien HERMENIER, FLUX, U. Utah & OASIS, INRIA/CNRS, U. Nice
Sophie DEMASSEY, TASC, INRIA/CNRS, Mines Nantes
Xavier LORCA, TASC, INRIA/CNRS, Mines Nantes
1

2. CONTEXT
datacenters and virtualization
2

3. DATACENTERS
interconnected servers
hosting
distributed applications
3

4. RESOURCES
server
capacities
application
requirements
4

5. VIRTUALIZATION
applications embedded in
Virtual Machines
colocated on any server
manipulable
1 datacenter
4 servers, 3 apps 5

6. ACTION
• stop/suspend
• launch/resume
• migrate live
has a known duration
consumes resources
impacts VM performance
6

7. ACTION
live migration
has a known duration
consumes resources
impacts VM performance
6

8. DYNAMIC SYSTEM
Applications Servers
• submission
• removal (complete, crash)
• requirement change (load
spikes)
• addition
• removal (power off, crash)
• availability change
7

9. DYNAMIC
RECONFIGURATION
8

10. RECONFIGURATION PLAN
migrate S1 - S2 migrate S4 - S1 launch S4
time
t=0
9

11. RECONFIGURATION
PROBLEM
given an initial configuration and new requirements:
• find a new viable configuration
• associate actions to VMs
• schedule the actions to resolve violations and dependencies
s.t every action is complete as early as possible
10

12. + USER REQUIREMENTS
Clients Administrators
• fault-tolerance w. replication
• performance w. isolation
• resource matchmaking
• maintenance
• security w. isolation
• shared resource control
to satisfy at any time during the reconfiguration
11

13. ENTROPY
an autonomous VM manager
12

14. PLAN MODULE
• reconfiguration problem: vector packing + cumulative
scheduling + side constraints (NP-hard)
Constraint
Programming
• trade-off fast+good
• composable online
• easily adaptable offline
13

15. ABSTRACTION
•1VM = 0 or 1 action
•min Σ end(A)
actions A
•full resource requirements
at transition
14

16. Compound CP model
Packing + Scheduling
decomposition degrades the objective and may result
in a feasible packing without reconfiguration plan
shared constraints resource+side
separated branching heuristic 1-packing 2-scheduling
15

17. Producer/Consumer
1 action = 2 tasks / no-wait
home-made cumulatives in every dimension
16

18. SIDE CONSTRAINTS
ban unary
fence unary
spread alldifferent + precedences
lonely disjoint
capacity gcc
among element
gather allequals
mostly spread nvalue
17

19. REPAIR
candidate VMs
fixed
heuristically
resource+placement constraints come with a new service:
return a feasible sub-configuration
18

20. EVALUATION
19

21. PROTOCOL
500 - 2,500 homogeneous servers
2,000 - 10,000 heterogeneous VMs
extra-large/high-memory EC2 instances
3-tiers HA web applications with replica
50% VM grow 30% uCPU
4% VM launch, 2% VM stop
1% server stop
20

22. LOAD
solving time 1,000 servers
70% = standard average consolidation ratio
21

23. SCALABILITY
solving time 5 VMs : 1 server
2,000 servers = standard capacity of a container
22

24. field to investigate for packing+scheduling
≠ usages to optimize: CPU, energy, revenue
side constraints important for users
Entropy: CP-based resource manager, fast,
scalable, generic, composable, flexible, easy to
maintain
http://entropy.gforge.inria.fr/
CONCLUSION
23

25. user constraint catalog
routing and application topology constraints
soft/explained user constraints
new results available:
http://sites.google.com/site/hermenierfabien/
PERSPECTIVES
24