Y ifei Yuan Anduo Wang Rajeev Alur Boon Thau Loo U niversity of Pennsylvania M otivation Managing network resources is the key computational problem in Data Centers Applying verificationsynthesis tool to network resource management ID: 464010
Download Presentation The PPT/PDF document "Automated Bandwidth Allocation Problems ..." is the property of its rightful owner. Permission is granted to download and print the materials on this web site for personal, non-commercial use only, and to display it on your personal computer provided you do not modify the materials and that you retain all copyright notices contained in the materials. By downloading content from our website, you accept the terms of this agreement.
Slide1
Automated Bandwidth Allocation Problems in Data Centers
Y
ifei
Yuan,
Anduo
Wang, Rajeev
Alur
, Boon
Thau
Loo
U
niversity of PennsylvaniaSlide2
Motivation
Managing network resources is the key computational problem in Data Centers.
Applying verification/synthesis tool to network resource management?
Benefits: exact solutions, correctness guaranteesChallenges: efficiencyThis work: bandwidth allocation by SAT/SMT solvers
1Slide3
Bandwidth Allocation Problem
2Slide4
Bandwidth Allocation Problem
1
G bps
600M bps
500M
bps
450M
bps
X
1
X
2
X
3
S
1
S
2
S
3
S
4
D
ata Center’s Network
10
G bps
10
G bps
2Slide5
Bandwidth Allocation Problem
1
G bps
600M bps
500M
bps
450M
bps
X
1
X
2
X
3
S
1
S
2
S
3
S
4
D
ata Center’s Network
V
1
V
2
V
3
400M
bps
400M bps
V
irtual Network
10
G bps
10
G bps
2Slide6
Bandwidth Allocation Problem
1
G bps
600M bps
500M
bps
450M
bps
X
1
X
2
X
3
S
1
S
2
S
3
S
4
D
ata Center’s Network
V
1
V
2
V
3
400M
bps
400M bps
V
irtual Network
10
G bps
10
G bps
2Slide7
B
andwidth Allocation Problem
1
G bps
600M bps
500M
bps
450M
bps
X
1
X
2
X
3
S
1
S
2
S
3
S
4
D
ata Center’s Network
V
1
V
2
V
3
400M
bps
400M bps
V
irtual Network
v
1
v
3
v
2
10
G bps
10
G bps
2Slide8
B
andwidth Allocation Problem
1
G bps
600M bps
500M
bps
450M
bps
X
1
X
2
X
3
S
1
S
2
S
3
S
4
D
ata Center’s Network
V
1
V
2
V
3
400M
bps
400M bps
V
irtual Network
v
1
v
3
v
2
10
G bps
10
G bps
2Slide9
BAP:
Facts
Complexity:
NP-complete: tree for physical network & virtual networkExisting heuristics:Pros: efficientCons: no guaranteeAlternative approach: SAT/SMT solving
3Slide10
SAT/SMT E
ncoding: A Glimpse
X(
v,s): VM v is mapped to server sY(l,e):
physical link
l
is reserved bandwidth virtual link
e
R(l,e,k
): physical link
l
is the
k-th edge on the routing path for virtual link eServer capacity:
∑v X(v,s) <
c(s), for every server sLink capacity:∑
e Y(l,e) < b(l), for every physical link
l
4Slide11
Abstraction and Refinement
O
bservation: Hierarchical physical network topology in data centers
TreeFat-treeIdea:Abstract physical network: small sizeRefine subgraphs
5Slide12
Abstraction
1
2
4
2
6Slide13
Abstraction
1
2
4
2
6Slide14
Abstraction
1
2
4
2
3
6
6Slide15
Abstraction
3
6
6Slide16
Abstraction
3
6
6Slide17
Refinement
1
2
4
2
3
6
7Slide18
Refinement
1
2
4
2
7Slide19
Evaluation: Set up
P
hysical network topology: tree with 200 servers:
200
20
4
4
4
4
8Slide20
Evaluation: Set up
V
irtual network topology: connected cliques
2
1
1
9Slide21
Evaluation: Set up
Experiment:
Run allocation algorithm
Keep mapping the VN to the PNStop when no more VN can be mapped10Slide22
Evaluation: Server Utilization
11Slide23
Evaluation: Link UtilizationSlide24
Evaluation: Running Time per VN
12Slide25
Summary
Alternative approach solving network resource allocation problem: using
SAT/SMT
solversAbstract&refinement for scalabilityStrength: optimal solutionWeakness: efficiencyPossible scenario: Optimal reallocation
13