An Opportunistic Resource Sharing and Topology-Aware

Mapping Framework for Virtual Networks

Sheng Zhanga, Zhuzhong Qiana, Jie Wub, and Sanglu Lua

aNanjing University bTemple University

INFOCOM 2012Orlando, FL

March 25 – 30, 20121

Network Virtualization⃝�Infrastructure provider (InP): physical/substrate network (SN)

⃝�Service provider (SP) purchases slices of resource (e.g., CPU,

bandwidth, memory) from the InP and then creates a

customized virtual network (VN) to offer value-added service

(e.g., content distribution, VoIP) to end users

2

Virtual Network Mapping⃝�VNM is to embed multiple VN requests with resource

constraints into a substrate network Different virtual nodes -> different substrate nodes VN requests arrive over time: first come, first serve

⃝�The objective is to maximize the revenue of InP, that is, maximize the utilization ratio of physical resources

VN request 1

VN request 2

Virtual Network MappingGiven a VN request and a substrate nerwork, the problem of determining whether the request can be embeded without any constraints violation is NP-hard[Andersen 2002]

4

Related Work⃝�Simulated Annealing: [Ricci et al. 2003][Zhang et

al. 2011]

⃝�Load Balancing: [Zhu & Ammar 2006]

Unlimited resources

⃝�Path Splitting: [Yu et al. 2008]

Multi-commodity flow problem

⃝�Location Constraints: [Chowdhury et al. 2009]

Integer Linear programming + determinstic/randomized rounding

⃝�Inter-domain mapping: [Chowdhury et al. 2010]

5

Motivation

6

⃝�It is difficult to predict the workload precisely

SP potentially target users all over the world

⃝�SPs often over-purchase physical resources

To cope with a peak workload on demand

unefficient

resource utilizatio

n

The ORSTA framework

7

1: Topology-aware node ranking (MCRank)

2: Macro level mapping - Greedy node-to-node mapping

- maximum first - Link-to-link mapping

- shortest path

3: Micro level assignment: for each substrate node and link, - Local time slot assignment

Step 1: Topology-Aware Node Ranking-Motivational Example

8

12 CPU, 8 Bandwidth

12 CPU, 2 Bandwidth

VN request 1

Topology-Aware Node Ranking-Basic Idea

9

PageRank: The importance of a web page is determined not only

by its own contents but also its neighbors’

Our observation:The importance of a substrate node is determined not

only by its own resource but also its neighbors’

Topology-Aware Node Ranking-Details

10

Actually, MCRank is the stationary distribution of a Markov chain

⃝�A node has a higher rank if it has more highly-

ranked neighbors

⃝�The more neighbors one node has, the less its

influence on their rankings

We prove the existence of MCRank, and also give an algorithm for calculating it. Please refer to paper for details.

Iterative effect

Step 2: Macro Level Mapping

11

⃝�Phase 1: node-to-node

Sort VN nodes according to their CPU requirements

Sort SN nodes according to their MCRank

Maximum first matching

⃝�Phase 2: link-to-link

shortest path

• y-z: G-H-D ?

G-F-E-D ?

k-shortest path

• multiple edges

VN request 1

Step 3: Micro Level Time Slot Assignment- Capture the fluctuation of workload⃝�Workload model

Basic part: always exists, its percentage is bwl

Variable part: each unit occurs with a probability, pwl, in each time slot

⃝�CPU busy time and network flow: expressed in time slots

proportional to the workload

12

Examples:Node “x”: basic 6 + variable 6The possible units needed: 6,7,8,…,12

bwl=0.5pwl=0.2

Step 3: Micro Level Time Slot Assignment

13

⃝�Only focus on a substrate link

Results can be applied to substrate nodes without any major

changes

⃝�Only focus on variable sub-traffic in a substrate link

For basic sub-traffic, we have no choice but to allocate the required

number of time slots

⃝�For variable sub-traffic

SHARE !

Step 3: Micro Level Time Slot Assignment- Tradeoff

14

⃝�When more than one variable sub-traffic occurs at

the same time slot, a collision happens.

⃝�To save time slots for upcoming requets

A slot is shared among, the more virtual links the better

⃝�To guarantee performance

A slot is shared among, the less virtual links the better

A tradeoff!

Step 3: Micro level Time Slot Assignment- Breaking the tradeoff

15

Given multiple variable sub-traffic and a collision threshold, find an assignment to minimize the slots used

Bin PackingFirst-fit

How to accelerate the calculation of collision probability? See paper.

Simulation Setup⃝�Performance metrics

Acceptance ratio: the higher, the better

Node/link utilization: the higher, the better

⃝�Algorithms in comparison

ORSTA: our entire framework

TA: only considers topology-awareness

ORS: only considers opportunistic resource sharing

Greedy: traditional greedy node and link mapping

16

Results: Comparison of algorithms

17

18

Results: Comparison of algorithms

19

Results: Comparison of algorithms

20

Results: Impacts of parameters

Conclusions⃝�We re-examined the virtual network embedding

problem from two novel aspects

Topology-awareness

Opportunistic resource sharing

⃝�We proposed a mapping framework, ORSTA,

which contains three main components

Topology-aware node ranking

Macro level mapping

Micro level time slot assignment

21

Thanks for your attention!

Q&A

http://cs.nju.edu.cn/dislab/

The Internet is a great success!

⃝�Information exchange

⃝�Applications support

⃝�Critical infrastructure

Like many successful technologiesthe Internet is suffering the adverse effects of

inertia

Internet Ossification

⃝�Multiple network domains with conflicting

interests

multilateral relationship? Difficult!

Deploy changes/updates? Global agreement!

⃝�The ever-expanding scope and scale of the

Internet’s use

security, routing stability, etc.

Flexibility + Diversity

Flexibility + Diversity 24

Simulation SetupSimilar settings to several existing studies

⃝�Substrate network

Topology: ANSNET/Arpanet

CPU & Bandwidth: [50,100], uniform

Collision threshold: 0.1

⃝�Virtual network

# of nodes: [2,10], uniform

Each pair of nodes connects with probability 0.5

Lifetime: 10 minutes, exponential

Arrivals: Possion process (0.2 minutes)25

Motivation 1: example

26

InP gets: 8$SP1 or SP2 pays: 4$

1$ for one unit per hour

InP gets: (3+0.1)*3=9.3$

SP1 or SP2 or SP3 pays: 3.1$

0.1$ for the shared unit per hour

Assumption: 4 units demand= 3 units (always needed) + 1 unit (needed with probability 0.1)

No Free Lunch! Collision may happen. (0.028 here)

Residual Resource Estimation

27

The residual room in a time slot is defined

as:

the maximal probability of a variable sub-

traffic that this slot can still accommodate.

The ORSTA Framework

28

Topology-Aware Node Ranking⃝�PageRank’s core idea

A page has a higher rank if it is pointed to by more highly-ranked pages

The more pages one page points to, the less its influence on their ranking is

⃝�MCRank

⃝�We prove that the Markov chain determined by P has a stationary distribution, i.e., MCRank. 29