a flow aggregation method based on end-to-end delay in · pdf fileofficer: a general...
TRANSCRIPT
A Flow Aggregation Method Based on End-to-End Delay
in SDNTakuya Kosugiyama†, Kazuki Tanabe†, Hiroki Nakayama‡,
Tsunemasa Hayashi‡ and Katsunori Yamaoka†
† Tokyo Institute of Technology, Japan‡ Bosco Technologies Inc., Japan
IEEE ICC 2017 CQRM“A Flow Aggregation Method Based on End-to-End Delay in SDN”
T. Kosugiyama, K. Tanabe, H. Nakayama, T. Hayashi, K. YamaokaOverview
BackgroundGoalModelingAlgorithmResultSummary
2
IEEE ICC 2017 CQRM“A Flow Aggregation Method Based on End-to-End Delay in SDN”
T. Kosugiyama, K. Tanabe, H. Nakayama, T. Hayashi, K. YamaokaWhat is SDN?
Software-Defined Networking (SDN)• Controllable via API
3
Application Plane
Control Plane
Data Plane
Business Application
Northbound API
Southbound API
Network ServiceSDNController
Network Device Network Device
Network DeviceNetwork Device
IEEE ICC 2017 CQRM“A Flow Aggregation Method Based on End-to-End Delay in SDN”
T. Kosugiyama, K. Tanabe, H. Nakayama, T. Hayashi, K. YamaokaProblem
Routing Mechanism in OpenFlow (OF)
4
1. Proactive 2. ReactiveOF Controller
OF Switch
① Configure rule
1 2
dst IP: A → Port: 2
③ Forwarding
1 2To: A To: A
③ Configure rule
dst IP: A → Port: 2② Request rule
④ Forwarding
② Packet comes ① Packet comes
[1] S.H. Yeganeh, A. Tootoonchian, and Y. Ganjali, “On scalability of software-defined networking,” IEEE Communications Magazine,vol.51, no.2, pp.136–141, Feb. 2013.
OF Controller
OF Switch
IEEE ICC 2017 CQRM“A Flow Aggregation Method Based on End-to-End Delay in SDN”
T. Kosugiyama, K. Tanabe, H. Nakayama, T. Hayashi, K. YamaokaProblem
Routing Mechanism in OpenFlow (OF)
5
1. Proactive 2. ReactiveOF Controller
OF Switch
① Configure rule
1 2
dst IP: A → Port: 2
③ Forwarding
1 2To: A To: A
③ Configure rule
dst IP: A → Port: 2② Request rule
④ Forwarding
② Packet comes ① Packet comes
• Rule update rate (1000 rules/s)• Rule capacity (10〜50 K) [1] SDN scaling issues
[1] S.H. Yeganeh, A. Tootoonchian, and Y. Ganjali, “On scalability of software-defined networking,” IEEE Communications Magazine,vol.51, no.2, pp.136–141, Feb. 2013.
OF Controller
OF Switch
IEEE ICC 2017 CQRM“A Flow Aggregation Method Based on End-to-End Delay in SDN”
T. Kosugiyama, K. Tanabe, H. Nakayama, T. Hayashi, K. YamaokaGoal
How to manage massive flows in SDN• Aggregate flows and reduce number of flows
• Rule aggregation method based on bandwidth [1][2]• Rule division method to satisfy capacity [3][4]
There are no works to aggregate flows based on End-to-End allowable delay• Target: IoT / M2M flow
6
[1] F. Giroire, J. Moulierac, and T. K. Phan. Optimizing rule placement in software-defined networks for energy-aware routing. In 2014 IEEE Global Communications Conference, pp. 2523–2529, Dec 2014. [2] Xuan Nam Nguyen, Damien Saucez, Chadi Barakat, and Thierry Turletti. OFFICER: A general optimization framework for OpenFlow rule allocation and endpoint policy enforcement. Proceedings - IEEE INFOCOM, Vol. 26, pp. 478–486, 2015. [3] Yossi Kanizo, David Hay, and Isaac Keslassy. Palette: Distributing tables in software-defined networks. Proceedings - IEEE INFOCOM, pp. 545–549, 2013. [4] Nanxi Kang, Zhenming Liu, Jennifer Rexford, and David Walker. Optimizing the ”One Big Switch” Abstraction in Software-Defined Networks. Conext’13, p. 17, 2013.
A flow aggregation method to minimize number of flowssatisfying allowable delay
IEEE ICC 2017 CQRM“A Flow Aggregation Method Based on End-to-End Delay in SDN”
T. Kosugiyama, K. Tanabe, H. Nakayama, T. Hayashi, K. YamaokaApproach
Basis of flow aggregation• Consider flows on same section as one flow
• Route by aggregated flow• ex) Full mesh flows (allowable delay = 80)
7
f1
f2
f’1
f’2
f12
Tunnel and labeling
(a) Network (b) Maximum aggregationnot satisfying allowable delay
2020
2040
20+20+20+40= 100 > 80
• flow: 3 → 1
Link cost
IEEE ICC 2017 CQRM“A Flow Aggregation Method Based on End-to-End Delay in SDN”
T. Kosugiyama, K. Tanabe, H. Nakayama, T. Hayashi, K. YamaokaApproach
Basis of flow aggregation• Consider flows on same section as one flow
• Route by aggregated flow• ex) Full mesh flows (allowable delay = 80)
8
f1
f2
f’1
f’2
f12
Tunnel and labeling
(a) Network (b) Maximum aggregationnot satisfying allowable delay
2020
2040
Link cost
(c) Maximum aggregationsatisfying allowable delay
IEEE ICC 2017 CQRM“A Flow Aggregation Method Based on End-to-End Delay in SDN”
T. Kosugiyama, K. Tanabe, H. Nakayama, T. Hayashi, K. YamaokaModel setting (1)
SDN as a directed graph• Node:• Link:
• Link cost:
• Flow: • Allowable cost:
• Relation between flow and link
9
IEEE ICC 2017 CQRM“A Flow Aggregation Method Based on End-to-End Delay in SDN”
T. Kosugiyama, K. Tanabe, H. Nakayama, T. Hayashi, K. YamaokaModel setting (2)
Constraints to route flow
10
There is only one flow from a source node and to a destination node
IEEE ICC 2017 CQRM“A Flow Aggregation Method Based on End-to-End Delay in SDN”
T. Kosugiyama, K. Tanabe, H. Nakayama, T. Hayashi, K. YamaokaModel setting (2)
Constraints to route flow
11
There is no flow to a source node and from a destination node
IEEE ICC 2017 CQRM“A Flow Aggregation Method Based on End-to-End Delay in SDN”
T. Kosugiyama, K. Tanabe, H. Nakayama, T. Hayashi, K. YamaokaModel setting (2)
Constraints to route flow
12
Number of incoming flows equals to number of outgoing flows on a node
IEEE ICC 2017 CQRM“A Flow Aggregation Method Based on End-to-End Delay in SDN”
T. Kosugiyama, K. Tanabe, H. Nakayama, T. Hayashi, K. YamaokaModel setting (2)
Constraints to route flow
13
A node is visited only one time
IEEE ICC 2017 CQRM“A Flow Aggregation Method Based on End-to-End Delay in SDN”
T. Kosugiyama, K. Tanabe, H. Nakayama, T. Hayashi, K. YamaokaModel setting (3)
Constraint of allowable cost
Aggregate flows• Flows on same link and same direction
• Objective: minimize number of flows
14
IEEE ICC 2017 CQRM“A Flow Aggregation Method Based on End-to-End Delay in SDN”
T. Kosugiyama, K. Tanabe, H. Nakayama, T. Hayashi, K. Yamaoka
Deformation••
•
••
Model setting (4)15
Apply to
0,1-Knapsack Problem⇒ NP-hard
y and z are vectors (0,…1,…,0) which correspond to the index of h and g
IEEE ICC 2017 CQRM“A Flow Aggregation Method Based on End-to-End Delay in SDN”
T. Kosugiyama, K. Tanabe, H. Nakayama, T. Hayashi, K. Yamaoka
Overview1. Ascending sort by (minimum cost - allowable cost) ⇒ Fs2. Select the first element of Fs⇒ f3. [Flow compotision]
• Compose route of f using existing flows4. [Flow aggregation]
• Aggregate flows in Fs into existing flows5. Return to 2. unless Fs is empty
Heuristics16
Flow composition Flow aggregationExisting flows
= flexibility of changing route
IEEE ICC 2017 CQRM“A Flow Aggregation Method Based on End-to-End Delay in SDN”
T. Kosugiyama, K. Tanabe, H. Nakayama, T. Hayashi, K. YamaokaFlow composition
Path composition by Best-First Search (BFS)• Evaluation value of node :
• : number of times existing rules are used• : minimum cost to reach from to
• Prioritize two nodes
• Time complexity:
17
IEEE ICC 2017 CQRM“A Flow Aggregation Method Based on End-to-End Delay in SDN”
T. Kosugiyama, K. Tanabe, H. Nakayama, T. Hayashi, K. Yamaoka
Flow composition by Best-First Search•
Flow composition: Ex.118
35
3
2
2
4
4
2
35
3
2
2
4
4
2
Dashed arrows: existing flows
{0,0}
{0,3}
{0,5}
Evaluation value: {t, c}t : number of times existing rules are usedc : minimum cost to reach from source to current
IEEE ICC 2017 CQRM“A Flow Aggregation Method Based on End-to-End Delay in SDN”
T. Kosugiyama, K. Tanabe, H. Nakayama, T. Hayashi, K. Yamaoka
Flow composition by Best-First Search•
Flow composition: Ex.119
35
3
2
2
4
4
2
35
3
2
2
4
4
2
Dashed arrows: existing flows
{0,0}
{0,3}
{0,5}
{1,5}
Evaluation value: {t, c}t : number of times existing rules are usedc : minimum cost to reach from source to current
IEEE ICC 2017 CQRM“A Flow Aggregation Method Based on End-to-End Delay in SDN”
T. Kosugiyama, K. Tanabe, H. Nakayama, T. Hayashi, K. Yamaoka
Flow composition by Best-First Search•
Flow composition: Ex.120
35
3
2
2
4
4
2
35
3
2
2
4
4
2
Dashed arrows: existing flows
{0,0}
{0,3}
{1,5}
{2,9}
Evaluation value: {t, c}t : number of times existing rules are usedc : minimum cost to reach from source to current
{1,8}
IEEE ICC 2017 CQRM“A Flow Aggregation Method Based on End-to-End Delay in SDN”
T. Kosugiyama, K. Tanabe, H. Nakayama, T. Hayashi, K. Yamaoka
Flow composition by Best-First Search•
Flow composition: Ex.121
35
3
2
2
4
4
2
35
3
2
2
4
4
2
Dashed arrows: existing flows
{0,0}
{0,3}
{1,5}
{2,9}
{2,13}≦
Evaluation value: {t, c}
Number of adding flows: 2
t : number of times existing rules are usedc : minimum cost to reach from source to current
{1,8}
IEEE ICC 2017 CQRM“A Flow Aggregation Method Based on End-to-End Delay in SDN”
T. Kosugiyama, K. Tanabe, H. Nakayama, T. Hayashi, K. Yamaoka
Flow composition by Best-First Search•
Flow composition: Ex.222
35
3
2
2
4
4
2
35
3
2
2
4
4
2
Dashed arrows: existing flows
{0,0}
{0,3}
{1,5}
{2,9}
{2,13} >
{1,8}
IEEE ICC 2017 CQRM“A Flow Aggregation Method Based on End-to-End Delay in SDN”
T. Kosugiyama, K. Tanabe, H. Nakayama, T. Hayashi, K. Yamaoka
Flow composition by Best-First Search•
Flow composition: Ex.223
35
3
2
2
4
4
2
35
3
2
2
4
4
2
Dashed arrows: existing flows
{0,0}
{0,3}
{1,5}
{2,9}{1,8}
IEEE ICC 2017 CQRM“A Flow Aggregation Method Based on End-to-End Delay in SDN”
T. Kosugiyama, K. Tanabe, H. Nakayama, T. Hayashi, K. Yamaoka
Flow composition by Best-First Search•
Flow composition: Ex.224
35
3
2
2
4
4
2
35
3
2
2
4
4
2
Dashed arrows: existing flows
{0,0}
{0,3}
{1,5}
{2,9}{1,8}
{1,10}≦
Number of adding flows: 3
IEEE ICC 2017 CQRM“A Flow Aggregation Method Based on End-to-End Delay in SDN”
T. Kosugiyama, K. Tanabe, H. Nakayama, T. Hayashi, K. Yamaoka
1. Ascending sort by (minimum cost - allowable cost) ⇒ Fs2. Select the first element of Fs⇒ f3. [Flow compotision]
• Compose route of f using existing flows4. [Flow aggregation]
• Aggregate flows in Fs into existing flows5. Return to 2. unless Fs is empty
Overview
Heuristics25
Flow composition Flow aggregationExisting flow
f1 f2
f3 (f1, f2)
IEEE ICC 2017 CQRM“A Flow Aggregation Method Based on End-to-End Delay in SDN”
T. Kosugiyama, K. Tanabe, H. Nakayama, T. Hayashi, K. YamaokaFlow aggregation
Delete flows in Fs over existing flows• Calculate flows can be reached within allowable
delay by Dijkstra’s algorithm• Time complexity:
26
f3 (f1’, f2’)
f1
f2 Flow aggregationf1’
f2’
f3 can be routed using part of f1 and f2
IEEE ICC 2017 CQRM“A Flow Aggregation Method Based on End-to-End Delay in SDN”
T. Kosugiyama, K. Tanabe, H. Nakayama, T. Hayashi, K. Yamaoka
Simulation settings
Simulation• Topology
• Model: ER / BA• Size: |V| = 50, 500
• Parameters• Link cost: [5, 15], discrete uniform distribution• alpha: ratio of allowable cost of flows to the maximum
value of the minimum delay of flows• Ex) min cost = 10, alpha = 2 ⇒ Cf = 20
• Comparison• Minimum cost / Minimum hop routing with aggreagtion
27
Table: topologies
IEEE ICC 2017 CQRM“A Flow Aggregation Method Based on End-to-End Delay in SDN”
T. Kosugiyama, K. Tanabe, H. Nakayama, T. Hayashi, K. Yamaoka
Performance evaluation (1)
Effect of allowable cost• |F| = 1000
28
Fig. 1 Effect of allowable cost of flows (Scalefree-small)
There is not much effect of alpha
IEEE ICC 2017 CQRM“A Flow Aggregation Method Based on End-to-End Delay in SDN”
T. Kosugiyama, K. Tanabe, H. Nakayama, T. Hayashi, K. Yamaoka
Performance evaluation (1)
Effect of allowable cost• |F| = 1000
29
Fig. 1 Effect of allowable cost of flows (Scalefree-small)
Flow composition selects first route→ No more effect of allowable delay
IEEE ICC 2017 CQRM“A Flow Aggregation Method Based on End-to-End Delay in SDN”
T. Kosugiyama, K. Tanabe, H. Nakayama, T. Hayashi, K. Yamaoka
Performance evaluation (2)
Effect of number of flows• α = 5
30
Fig. 2 Random-large Fig. 3 Scalefree-large
HIGH link density→ more effective
IEEE ICC 2017 CQRM“A Flow Aggregation Method Based on End-to-End Delay in SDN”
T. Kosugiyama, K. Tanabe, H. Nakayama, T. Hayashi, K. YamaokaSummary & Future work
31
We propose a flow aggregation method to minimize number of flows satisfying End-to-End delaySimulation in several topologies• Higher performance than simple comparison method• Flexibility of changing route is importantFuture work• Expand our model more realistic• Bypassing on weak topology