B99705021 資管三 李奕德
http://ppt.cc/41rH







Introduction
Background
Virtual machine placement
Algorithm
Algorithm evaluation
Result
Discussion and future work
Scalability issue
 Aim to solve different problem
- Dcell, Bcube, PortLand, VL2……
 No thinking of traffic issue
- high traffic from end to end


1.
2.
3.

three character of all traffic
average pairwise traffic rate & end-to-end
cost has low correlation
Uneven between VMs
Stays almost the same
Traffic-aware placement may be beneficial






Traffic-aware VM Placement Problem
(TVMPP)
given: traffic matrix , cost matrix
Goal: minimize cost
Cost can be: Total switch used/Compute Time
An algorithm that solve the NP-hard problem
Architecture difference
NP: by nondeterministic algorithms in
polynomial time
 nondeterministic
-Every “guess by hunch” is right


at least as hard as the hardest problems in NP







Introduction
Background
Virtual machine placement
Algorithm
Algorithm evaluation
Result
Discussion and future work







Data set I :
IBM Global Services’ data warehouse
About 17000 virtual machines
Data set II:
Server cluster
About Hundreds of virtual machines
round-trip latency measurement at 68 VM

Uneven between VMs

80% of VM’s traffic < 800kb/sec
4% of VM’s traffic > 8mb/sec


Stays almost the same

Low correlation between average pairwise
traffic rate & end-to-end cost

Correlation : -0.32

Old style

VL2

Portland

Bcube







Introduction
Background
Virtual machine placement
Algorithm
Algorithm evaluation
Result
Discussion and future work




n VM to assign
n slot for VM
static and single-path routing
Cost and traffic matrix from historical data

Cost 
 D C       e g  
i , j 1,...,n



ij
i
j
i 1,...,n
i
i
is equivalent of finding


min tr DX T C T X  eX T g T
X 
Dummy VM is assigned when no. slot > no.
VM






Quadratic Assignment Problem (NP-hard)
Impossible to find optimality when size > 15
TVMPP is a special case of QAP
reduction from Balanced Minimum K-cut
Problem (BMKP)
BMKP: extended problem from the Minimum
Bisection Problem (MBP)
BMKP & MBP are NP-hard







Introduction
Background
Virtual machine placement
Algorithm
Algorithm evaluation
Result
Discussion and future work






approximation algorithm Cluster-and-Cut
Divide VM into VM cluster
Divide slot into slot cluster
Put VM cluster into slot cluster
A smaller problem
Feasible when size is sufficient small




Complexity determine by SlotClustering and
VMMinKcut
Slotclustering: O(nk)
VMMinKcut: O(n4)
Total complexity = O(n4)







Introduction
Background
Virtual machine placement
Algorithm
Algorithm evaluation
Result
Discussion and future work






Cluster and cut VS. other benchmark
algorithms
Local Optimal Pairwise Interchange (LOPI)
Simulated Annealing (SA)
hybrid traffic model
Gravity model
compute the GLB for each settings







Introduction
Background
Virtual machine placement
Algorithm
Algorithm evaluation
Result
Discussion and future work

Cost matrix

Compare with random assign

Traffic is assumed to be in normal distribution
Variance is change to show difference

Different architecture & variance affect result



View as VM cluster
GLB prediction

GLB prediction VS. optimal solution
Thing that brings better performance:
- bigger variance
- smaller cluster (less VM in a group)
- Architecture difference
(generally) Bcube > tree > fat-tree > VL2
 Good scenario: multiple service in a data
center
 Bad scenario: single service / map-reduce








Introduction
Background
Virtual machine placement
Algorithm
Algorithm evaluation
Result
Discussion and future work


Dynamic VM placement
Other VM placement with different goal
Thank you for your attention