• No results found

DESIGN AND ANALYSIS OF TECHNIQUES FOR MAPPING VIRTUAL NETWORKS TO SOFTWARE- DEFINED NETWORK SUBSTRATES

N/A
N/A
Protected

Academic year: 2021

Share "DESIGN AND ANALYSIS OF TECHNIQUES FOR MAPPING VIRTUAL NETWORKS TO SOFTWARE- DEFINED NETWORK SUBSTRATES"

Copied!
17
0
0

Loading.... (view fulltext now)

Full text

(1)

DESIGN AND ANALYSIS OF TECHNIQUES FOR

MAPPING VIRTUAL NETWORKS TO

SOFTWARE-DEFINED NETWORK SUBSTRATES

Tran Song Dat Phuc - Uyanga

Department of Computer Science and Engineering

SeoulTech 2014

(2)

Table of Contents

Introduction

Related work

Model and problem statement

(3)

Introduction

Software-defined networking (SDN) has emerged as a powerful

approach to improve the customizability and flexibility of

networks.

By separating the control plane from the data plane, SDN

facilitates network experimentation and allows for optimizing

switching/routing policies.

The forwarding planes in SDN are managed by remote

processes called controllers.

(4)

Introduction

There are some solutions for virtualization in SDN infrastructure, the

most popular are OpenFlow, FlowVisor and FlowN.

With virtualization, slices represent for networks in SDN. Each slice

has a slice policy defining its resources and the controller associated

with it.

Slices of the network can be used for many different purposes

depending on what the owners are trying to accomplish. Services or

experiments that run on these slices may display a wide variety of

resource requirements.

A slice along with its controller as a virtual network (VN) in SDN. As

VNs get bigger and the number of VNs increases, resource contention

may become a problem.

(5)

Introduction

Embedding VNs within a network virtualization environment is an

important problem (substrate network providers are decoupled from

VN providers that deploy and operate the VNs), because suboptimal

mappings can cause bad performance and/or higher operating cost.

Solutions that strive for well-balanced and resource-efficient VN

mappings have been proposed by researchers.

This study designs embedding techniques for VNs in the SDN

environment along with two goals : balancing the load on substrate

nodes and links, and maintaining low delay between controllers and

switches in all VNs.

(6)

Related work

SDN is a suitable platform for network virtualization and researchers

have been working on ways to provide virtualization to enable the

coexistence of multiple VNs in the SDN environment.

OpenFlow has been extensively used as a uniform interface between

the control and data planes in SDN.

FlowVisor, a special purpose OpenFlow controller that can create

slices of network resources and place each slice under the control of a

different OpenFlow controller.

FlowN utilizes virtualization to run a modified version of the

controller and also maps API calls between the physical and virtual

networks, but it does not employ a separate controller for each VN.

(7)
(8)

Related work

VN embedding with node and link constraints is a complex problems.

Each VN on an SDN-based substrate possesses its own controller, and

there are requirements and considerations that come with the

existence of this controller.

The controller is responsible for organizing the operation of the VN

by sending routing updates, traffic engineering policies etc. and needs

to react quickly to faults in the network.

The controller also must be able to communicate effectively to all the

switches that are part of the VN, so any VN embedding effort needs

to make sure all controller-to-switch channels avoid congestion and

high delays.

The ease of customizing packet routes in SDN presents an additional

degree of freedom in VN embedding.

(9)

Related work

• The distinctions of the SDN environment (such as the centrality and the importance of the controller, and differences in the virtualization technology) make VN embedding become a new challenge.

• The mapping of the virtual components to substrate components and the placement of the controllers are both important variables influencing the performance of the VNs in such a system. Efficient utilization of networks resources, low risk of congestion, reliability and fast response are all desirable properties.

• With those problems, this study designs a stress-balancing VN assignment algorithm to consider the stress-balancing on the substrate network, one of two VN embedding objectives.

• Another is minimizing the average and maximum delays from the controller to the switches.

• The placement of controllers at predetermined location or free, the topology of each VN, as well as the mapping of that topology to the SDN substrate is also determined.

(10)

Model and problem statement

SDN Resources

Virtualization in SDNs requires a mechanism to share network

resources among multiple slices.

FlowVisor presents some essential network resources:

• Switch CPU: Sharing power between slices at switches is considering by two main tasks performed for each slice: generating new flow messages to be sent to the corresponding controller, and handling controller requests regarding the slice. CPU power demands at a switch increase with the number of VNs and links at mapped to it/traversing it.

(11)

Model and problem statement

• Bandwidth: Each slice has its own queue at each port, and these queues are serviced according to the resource allocation policy. Balancing the numbers (the number of virtual links that share a substrate link and the number of controller-to-switch connections that must go over this link) across the substrate is useful to avoid potential hot spots and reduce the possibility of congestion.

• Flow entry space: The number of flow entries that can be used by each slice is also limited at each switch. When a controller is over its limit, it is not allowed to insert any new rules at the switch. Consider the number of virtual links traversing a substrate node as part of the load on that node.

(12)

Model and problem statement

VN embedding problem

Assume underlying SDN infrastructure as the network substrate and

model it as a graph (V, E), where V is set of substrate nodes and E is

set of substrate links.

While there are many VNs sharing the underlying SDN infrastructure,

the two important problems are determined:

(a) deciding which switches to include in a slice.

(b) how to configure the routing within the slice.

Focus on controller placement and VN embedding.

• Consider two options for each VN: the VN embedding with

fixed-location controller (VNE-F) and the VN embedding with adjustable-location controller (VNE-A).

(13)

Model and problem statement

• With a substrate node (the number of virtual nodes) and a substrate link

(the number of virtual links), we concentrate on the stress (as a measure of VN embedding load) through two definitions : node stress and link stress.

• For node stress, we take into account the additional CPU power and

flow entry space load that comes from virtual links that traverse a node.

Assume SN (v) be the stress of a substrate node v V, then:

where nN (v) is the number of virtual nodes assigned to it, nL (v) is the number of virtual links traversing it, α and β are two positive parameters.

(14)

Model and problem statement

• For link stress, we add a component for the total load due to all the

controller-to-switch communications going over a particular link.

• Total link stress is the sum of data traffic stress and control traffic

stress.

Assume SL (e) be the stress of substrate link e E, then:

where Ci (e): the number of controller-to-switch connections traversing link e in the ith VN out of numV - total VNs in the system,

0 ≤ γi ≤ 1: each controller-to-switch connection contributes γi to the stress of each link on the path.

(15)

Model and problem statement

Let Slidei be the ith slice sharing a network of switches. The virtual topology

of this slice, VTi , defines the nodes and links included in this slice. VTi is described as a graph (V’i , E’i), where V’i is the set of virtual nodes, and E’i is the set of virtual links. This slice, along with its controller Ci , constitute virtual network Ni .

Assume the virtual topology of all numV VNs, and the location of each Ci is fixed or adjustable, the VN embedding must be designed to organize all VNs in such a way that:

(a) minimize the maximum stress on a switch or link while keeping all controller-to-switch delays under a threshold R

(b) minimize the average controller-to-switch delay for each Ni while keeping the maximum stress under a threshold T.

(16)

Conclusion

This research presented the virtual network (VN) embebding

problems in SDN environment. Thus, it proposed a design of

new embedding techniques in an effort to consider VN

embebding and SDN controller placement together, that focus

on two main objectives:

Minimizing the controller-to-switch delay

Balancing the load on substrate nodes and links.

From this research, the further study can be developed related to

the VN reconfiguration in SDN, the controller placement,

virtual node mapping and virtual link mapping stages.

(17)

References

Mehmet Demirci, Mostafa Ammar, “Design and analysis

of techniques for mapping virtual networks to

software-defined network substrates”, Comput. Commun. (2014).

References

Related documents

Let them know that this kind of change is normal and provide them with resources, such as brochures, books or videos that may help them support your child during this

A veteran of the marketing technology industry with more than two decades of experience, he has previously served in senior executive positions at InQuira (acquired by Oracle),

Double functionalization of the DHA skeleton has been achieved applying the bromination/elimination protocol on 3-substituted DHA 15, 30 7-substituted DHA 17 30 and on a

four-way, crossover trial comparing the 24-h FEV 1 profile for once- daily versus twice-daily treatment with olodaterol, a novel long-acting β 2 -agonist, in patients with

Public Health England 5 requires NHS Health Check commissioners to provide high- quality training to HCPs conducting NHS Health Checks and needs to consider ways of meeting the

Secondary outcomes were the total number of days of antibiotic administration for each infection category recorded in care home medical records (urinary tract infec-

The data loggers to measure temperature and humidity were placed in the drying chamber, on the metal collector and outside the dryer.. At three intervals during the day, the weight

3 Attack Model C.: Malicious nodes launch DoR attacks to reduce the target nodes’ reputation and cause them to be evicted from their TGrps: Although these attacks can be successful,