Agile and Economic Media-centric Service Realization over Software-Defined Infrastructure

JongWon Kim*, Taeheum Na, Aris Cahyadi Risdianto, ByungRae Cha, Sun Park

Gwangju Institute of Science & Technology (GIST), Gwangju, Korea E-mail: jongwon@gist.ac.kr, Tel: +82-62-715-2219

Abstract — The lifecycle management of service realization is very challenging. With virtualized playgrounds over Future Internet testbeds, the lifecycle experiments could be easily exercised so that all tasks and responsibilities are well-defined for entire experiment stages among developers and operators. Also, the dynamic provisioning of hyper-convergent compute/networking/storage resources is appropriately streamlined with the experiment lifecycle. In this paper, by considering these issues, we discuss the agile and economic realization of automated media-centric experiments over OF@TEIN (OpenFlow @ Trans-Eurasian Information Network) SDI (Software-Defined Infrastructure).

I. INTRODUCTION

To match the ever-increasing demands for diversified media-centric services, Future Internet testbed projects are evolving from isolated small-scale experiments towards massive and automated realization of multi-purpose experiments [1]. In this paper, we focus on integrated futuristic testbeds for testing and staging experiments, where various hardware boxes and software tools are combined employing DevOps (Development and Operation) automation [2]. As evidenced in world-wide Future Internet testbed efforts, repeatable experiments over testbeds are now maturing to cover realistic large-scale simulations on theoretic hypothesis, experimental verifications of software-based functions, and real-world performance benchmarking for production services. Also, the on-going convergence of ICT (Information and Communications Technology) infrastructure is heavily tied with the rapid adoption of SDN (Software Defined Networking), NFV (network functions virtualization), and CC (Cloud Computing). This kind of fundamental changes will enable continuous integration and delivery for agile and economic service realization by leveraging upcoming SDI (Software-Defined Infrastructure) based on SDN/NFV/Cloud infrastructure convergence.

As illustrated in Fig. 1, the diverse services can be easily

created by flexibly utilizing the hyper-convergent (i.e., compute/networking/storage integrated) resources from the provisioned resource pools. The developers are playing with virtualized playgrounds customized for themselves, where the required orchestration is made to support zero-touch installation/configuration (i.e., provisioning), flexible controls,

and instantaneous visibility. Note that the deeply-programmable and virtualized resources, leveraging SDN/NFV/Cloud technology, are the key enabler for upcoming SDI that could be effectively shared among all service developers employing template-based DevOps automation.

Com pute

Storage

N etw orking X

Zero-touchConfiguration

FlexibleControl

(forw arding,…)

Instant V isibility

DevOps

Fig. 1: Service realization concept with DevOps-based configuration/control/visibility automation.

Thus, in this paper, by taking the vision of building user-

customizable virtualized playgrounds, we will review and discuss the on-going development of OF@TEIN project. With an internationally distributed SDN-enabled SmartX Racks, the OF@TEIN project, initiated from 2012, is focusing on how we could support the agile and economic realization of automated diverse experiments. More specifically, we discuss the realization of automated media-centric experiments over OF@TEIN SDI.

II. OF@TEIN TESTBED INFRASTRUCTURE

A. OF@TEIN SDN-enabled Testbed For the successful verification of Future Internet research

ideas, the construction and federated operation of SDN-enabled testbeds is highly important. Since 2008, various projects on Future Internet testbeds, such as US GENI (Global Environment for Network Innovations), EU FIRE (Future Internet Research & Experimentation), and Japan NWGN (New-Generation Network), have started along with the wide adoption for OpenFlow-based SDN technology [1]. Thus, in

978-616-361-823-8 © 2014 APSIPA APSIPA 2014

July 2012, to promote the easy construction and wide expansion of SDN-enabled testbeds, we launched OF@TEIN project. With OF@TEIN, we targeted to gradually build and operate an OpenFlow-based SDN testbed over TEIN4 (Trans-Eurasia Information Network 4). This OF@TEIN collaboration project is being carried out by a consortium of Korean universities and international collaboration sites, led by GIST, Korea [3].

NIA(Seoul)

Indonesia

Malaysia (UM)

Vietnam

Philippines

Networked Tiled Display

SmartX Box (Type B+)

EU (SmartFIRE)

Japan or USA

OF@TEINOpenFlow Switch

Exp. Node (with HD camera)Exp. Node (traffic generator)Exp. Node

OpenFlow Production Switch

OpenFlow FlowVisorOpenFlow Controller

OF@TEIN PortalOF@TEIN SDN Tools

SmartX Rack (Box)

Jeju (Jeju)

Thailand

GIST (Gwangju)

VoD

Korea U (Seoul)

Postech(Pohang)

OF@KOREN

SmartXBox

(Type C)

Malaysia (MYREN)

Pakistan

Fig. 2: OF@TEIN testbed infrastructure (2012~2013).

More specifically, unique SmartX Racks are deployed to

promote the international collaboration with TEIN NRENs (National Research & Education Networks), resulting in an OpenFlow-based SDN testbed over 12 international sites in 7 countries (Korea, Indonesia, Malaysia, Thailand, Vietnam, Philippines, and Pakistan). The overview diagram for deployed OF@TEIN infrastructure is presented in Fig. 2, where the red-/blue-colored icons represent Type B+ and C SmartX Racks, respectively. OF@TEIN SmartX Racks are deployed to simultaneously support hyper-convergent (i.e., computing and networking) resources. They are coordinated by two separate planes: One for control/management and the other for datapath forwarding.

B. SmartX Racks and Operation Nodes

As shown in Fig. 3, we utilize SmartX Racks and Operation Nodes to operate OF@TEIN SDN-enabled testbed. We designed several types (Type A, Type B, Type B+, Type C) of SmartX Racks to provide varying combinations of resources. For example, each SmartX Rack (Type B) consists of four devices such as Management & Worker node, Capsulator node, OpenFlow switch, and Power management device. Especially, as shown in Fig. 3, OpenFlow-switched tunneling is adopted to enable the multi-international connections among SmartX Racks [4]. That is, OpenFlow switches (that connects the worker virtual machines) are linked by the L2-GRE tunneling of OpenFlow-aware Capsulators. From early 2014, we began to minimize the number of unique nodes in a SmartX Rack and shifted to SmartX Boxes by leveraging the virtualization power of hyper-convergent resources. For this, through Chef DevOps

automation software [5], all SmartX Racks are now upgraded by remotely managing power control and installation/configuration procedures.

OF@TEIN Underlay Network

Worker VM #1

Monitoring Agent

OpenFlowAgent

SmartX-Rack AgentMediaX-VT

Agent

Worker VM #2

Worker VM #3

Ope

n vS

witc

h

NARINET OF Switch & Capsulator Node

Management Switch

Worker VM #1

MonitoringAgent

Worker VM #2

Worker VM #3

Rem ote Pow er M ng.

NF/OVS CapsulatorNode

Management Plane Tunneled Data Plane Worker VMData Plane Management VM

OF@TEIN Portal Server

SmartX Rack (Type A*)SmartX Rack (Type B)

VM

#2VM

#1

VM

#2

OpenFlowAgent

SmartX-Rack Agent

MediaX-VT Agent

VM

#1

OF@TEIN System / Network Monitoring Node

System Monitoring

NetworkMonitoring

FloodLight

OF@TEIN SDN Tool NodeNOX

OF@TEIN OpenFlowManagement Node

FlowSpaceMonitoringFlowVisor

Dataplane OF SwitchD ata Manag.

Management VM (SmartX-

Rack / OpenStack/ Monitoring

Agents)

Worker VM #1

Worker VM #2

Worker VM #3

Ope

n

vSw

itch

Management Switch

Rem ote Pow er M anagem ent

SmartX Rack (Type B+/C)

Storage Manage

ment

TEIN-HK

APAN-JP

Vietnam

Philippines

Indonesia

Thailand

Malaysia

TEIN-JP

TEIN-SG

Parkistan

KOREN

Alarm

Netflow /cflow

weather map

traffic

OF@TEIN BOXMaster Nodes

Installer Coordinator

Fig. 3: OF@TEIN testbed: SmartX Racks and Operation Nodes.

C. Resources, Functions, and Service Composition An effective and differentiated framework for futuristic

service realization can be understood with three main ingredients such as resources, functions (in the sense of compute/networking/storage services), and composed services [9]. In order to convert the creative minds of developers into reality, we need to overcome the old-fashioned and closed way for service realization. By utilizing the hyper-convergent SmartX Rack resources, the developers can now create the required service functions (depicted as the color pens in Fig. 1), which will be consumed via RESTful APIs. For this, they need to dynamically provision all available resources so that the service-composition experiment could be automated. In short, with OF@TEIN, we are targeting an integrated SDI-style testbed that comprises all hardware boxes and software.

The developers should create their own services by

utilizing the programmability of hyper-convergent SmartX Rack resources. To the individual developer, the efficiency and flexibility of on-demand resource provisioning is highly important, which frees them from the constrained containers (e.g., denoted as boxes) of resources. With the supported virtualization of resources, a number of developers can simultaneously realize their own services without interfering with others. Thus, with OF@TEIN, we conceptualize the idealistic shaping of futuristic resources with SmartX Racks (or Boxes), which represents the configurable containers of various form-factor resources. That is, the scaled-out provisioning of resource pools is made to improve the economic incentives while the selective sets of heterogeneous resources are supported to enhance the functional diversity. In summary, we should leverage both programmable and virtualized (albeit hyper-convergent) resource pools for diversified service realization.

III. OF@TEIN LIFECYCLE EXPERIMENTS AND AUTOMATION

A. Lifecycle Experiments

The lifecycle of service realization experiment is composed of multiple stages, where tasks and responsibilities are well-defined between developers and operators. We attempt to prototype media-centric experiments over an OF@TEIN SDN-enabled testbed while paying attention to the automated resource provisioning and experiment execution [6]. By dividing into several experiment stages for designing, provisioning, executing, monitoring, and finishing, we define OF@TEIN lifecycle experiments to assist the operators in providing the needed resources and software tools for the developers. At the same time, it can also help the developers to understand the different stages of repeatable lifecycle experiments.

More specifically for OF@TEIN, we support the operators

to monitor, check, and recover the available resources before and during lifecycle experiments. We also support the developers to play with other tools/scripts for their own experiments. For this, we extend our previous ‘workflow-based control of experiment’ work [7], by automating resource provisioning (i.e., installation/configuration) and experiment execution via easy-to-use scripting. As a result, we can provide fast and efficient realization of lifecycle experiments over OF@TEIN. As shown in Fig. 4, a preliminary design of OF@TEIN lifecycle experiment is thus proposed by dividing resource provisioning and experiment execution stages and by adding detailed tasks with specific workflow for each stage. Fig. 4 shows the design stage to prepare an experiment scenario and define required tasks before executing the experiment. The next stage will be the provisioning stage where the operators need to provide experiment resources for the developers. Experiment execution stage is then defined for the developers to fully control the experiment by using simple tools/scripts, which is followed by the finish stage to release allocated resources.

Fig. 4: OF@TEIN preliminary design of lifecycle

experiments.

B. DevOps Automated Resource Provisioning.

This lifecycle experiment design includes privilege and responsibility distinction between the operators and the developers. Almost all stages involve both the operators and the developers. However, they cover different tasks and thus the associated workflows depend on the level of tasks (as well as privileges). That is, both developers and operators should build DevOps relationship, where the developers are responsible to rapidly develop their services on the top of testbed infrastructure while the operators are efficiently managing the testbed infrastructure.

Physical Network

Overlay Tunnels

VMVMVM

L2

A Virtual Playground

L2

L3

VM

Configuration Configuration

ControlVisibility

OpenvSwitch

VM VM VM

Neutron Swift Cinder

NOVA Glance

SmartXBox

(Type C)

SSD/HDDs

CPUs/GPUs

Box/Function/Topology Templates

Box

Coordinator

Computing

Default

FunctCCNX

Web

Server

Traffic generator

Operator

Topology

VM Images

NodeGraphs

Fig. 5: OF@TEIN virtual playground automated installation & configuration

with templates.

As mentioned above and illustrated in Fig. 5, in order to have faster execution time and provide automatic execution, the resource provisioning adopts the DevOps combination of Chef [5] installation/configuration and script-driven experiment execution [6]. Chef is very effective in repeating the same installation/configuration based on Ruby-written recipes from cookbooks, which can automatically transform SmartX Rack nodes into a ready-to-be-utilized resource pool. Like this, the above combination supports flexibility and efficiency for agile and economic realization of media-centric service experiments by providing automated resource provisioning and recovery.

C. Media Distribution Experiments

With the dynamically-provisioned SmartX Rack resources, we now prepare the required functions and inter-connect them to construct the desired service chaining for service realization. For example, this task can be handled by an experiment preparation script that first checks IP subnet and FlowSpace (tied with VLAN id for each developer) resource allocation, networking resource (OpenFlow switches and tunnels) registration to the FlowVisor, and computing resource (virtual machines) creation. The IP subnets, as the parameters for developer

experiments, are allocated by the operators. Then, the experiment execution script will start ping tests among all virtual machine pairs, which can verify the computing-side resource availability. We finally create an output that details about automated resource checking. Similarly, we provide experiment visibility with so-called OF@TEIN Experimenter UI. This Java-based GUI developed in [8] shows the allocated resources and active traffic flows in a single UI view.

Fig. 6: OF@TEIN media distribution experiment.

IV. CONCLUSIONS

With the advancement of SDN/NFV/Cloud-based SDI, the differentiated support - equipped with hyper-convergent resources - has become the key requirement of future ICT infrastructure. To verify easy and fast lifecycle experiments over the SDI-oriented testbeds, several automated provisioning and execution for OF@TEIN media distribution lifecycle experiments is performed. First, we presented how to understand the creation of futuristic agile and economic service realization over programmable and virtualized resource pools. We then describe how to facilitate the buildup of futuristic ICT infrastructure to assist the proposed service realization vision. In near future, by fully utilizing DevOps-based automation tools, we plan to continue diversified lifecycle experiments to further refine the task automation of all stages for lifecycle experiments.

ACKNOWLEDGMENT

This work was supported by one of KOREN projects of National Information Society Agency (2014-Smart-WI29). Also, this work was supported in part by the Industrial Strategic Technology Development Program (10047577, Development of Home Gateway for the SDN Based Intrusion Response) funded By the Ministry of Science, ICT and Future Planning. (MSIP, Korea).

REFERENCES

[1] M. Berman, J. S. Chase, L. Landweber, A. Nakao, M. Ott, D. Raychaudhuri, R. Ricci, and I. Seskar, “GENI: A federated testbed for innovative network experiments,” Computer Networks, vol. 61, pp. 5-23, 2014.

[2] M. Hüttermann, DevOps for developers. vol. 1. New York, USA, Apress, 2012.

[3] J. Kim et al., “OF@TEIN: An OpenFlow-enabled SDN testbed over international SmartX Rack sites,” in Proc. APAN - NRW, Daejeon, Korea, Aug. 2013.

[4] N. Kim, J. Kim, C. Heermann, and I. Baldine, “Interconnecting international network substrates for networking experiments,” in Proc. Tridentcom 2011, Shanghai, China, Apr. 2011.

[5] Opscode, Chef Documentation. http://docs.opscode.com/. [6] A.C. Risdianto, T. Na, and J. Kim, “Running lifecycle

experiments over SDN-enabled OF@TEIN testbed,” in Proc. IEEE ICCE, Danang, Vietnam, July 2014.

[7] S.W. Han, N. Kim, and J. Kim, “An experimental service composition tool for media-centric networked applications,” Computer Networks, vol. 63, pp. 188-204, 2014.

[8] S. Shin and J. Kim, “Towards service-aware flow visualization over OpenFlow-based programmable networks,” in Proc. APAN - NRW, Delhi, India, Aug. 2011.

[9] J. Kim, “Realizing futuristic service composition with programmable and virtualized pools of unified resources,” in Proc. Int. Conf. on Future Internet Technologies (CFI 2013), Beijing, China, June 2013.