Graph-based Campaign Amplification in Telecom Cloud

M. Saravanan

, Sandeep Akhouri

and Loganath Thamizharasu

Ericsson R&D, Ericsson India Global Services, Chennai, India

College of Engineering Guindy, Anna University, Chennai, India

Keywords: Telecom Cloud, Campaign Management, NoSQL Graph Database, Map Reduce Framework.

Abstract: Majority of telecom operators are making a transition from a monolithic, stove-pipe approach of creating

services to a more flexible architecture that provides them agility to rapidly develop and deploy services.

New revenue streams require an ability to rapidly identify and target dynamic shifts in traffic patterns and

subscriber behaviour. As subscriber behaviour morphs with plans, promotions, devices, location and time,

this presents challenges and opportunities for an operator to create and launch targeted campaigns. The

enormous volume of data being generated requires a scalable platform for processing massive xDR (eg. Call

Detail Records). This paper proposes graph databases in a telecom cloud environment for quickly

identifying trends, isolating a targeted subscriber base and rapidly launching campaigns. We also highlight

the limitations of a conventional relational database in terms of capturing complex relationships as

compared to a NoSQL graph database and the benefits of automatic provisioning and deployment in the

cloud environment.

1 INTRODUCTION

As mobile penetration worldwide reaches near

saturation level, telecom operators are looking

beyond net additions of subscribers for revenue

growth. Improvements in operational efficiency

leading to increased consolidation and managed

services will help reduce costs. However, the growth

in revenues will be driven by innovative service

offerings and service differentiation. Tiered data

plans and Quality of Service (QoS) offerings are

steps in this direction. Monetizing new revenue

streams and leveraging partnerships will require

operators to significantly improve subscriber

intelligence.

We explore generating dynamic campaigns that

can be launched in parallel. A BigData platform in a

hybrid cloud environment provides ability to process

large volumes of data while leveraging the

advantages inherent with cloud platform (Zhang et

al., 2010). To enhance the process, we have used

another upcoming technology - graph databases in a

highly scalable Hadoop (Tom, 2009) framework.

Sophisticated queries were efficiently modelled

using a NoSQL graph database for generating

various types of campaigns in a telecom cloud

environment (Leavitt, 2010). We also explored the

graph indices available with the graph database for

generating the node-level summary.

The main contributions of this paper are: (i) a

newly developed software framework for modeling

cloud computing environments (ii) An end-to-end

cloud network architecture that supports the huge

telecom volume and various services and (iii) It

exposes powerful graph-based features that could

easily be extended for modeling custom cloud

computing environments (like automatic campaigns)

and other applications.

2 BACKGROUND

Cloud computing is a promising paradigm for the

provisioning of IT services in any new applications

(Paul and Dan, 2010). Recent advances in cloud

computing, BigData platforms such as Hadoop and

NoSQL technologies present a unique opportunity

for telecom operators to improve upon their

subscriber intelligence. Call Detail Records

(CDRs/xDRs) contains the details of the calls made

by the customers. The data related to the customers

interest added to the existing dataset. This synthetic

data is added for launching targeted campaigns In

this paper, we use HDFS and MapReduce

195

Saravanan M., Akhouri S. and Thamizharasu L..

Graph-based Campaign Ampliﬁcation in Telecom Cloud.

DOI: 10.5220/0004050801950198

In Proceedings of the International Conference on Data Technologies and Applications (DATA-2012), pages 195-198

ISBN: 978-989-8565-18-1

 2012 SCITEPRESS (Science and Technology Publications, Lda.)

framework to develop a reliable and high

performance platform for processing, aggregation

and storage of CDRs. We leverage the automatic

provisioing and utility pay-as-you-use model

provided by a cloud computing (Armbrust et al.,

2010). We experimented with several graph

databases and chose InfiniteGraph (Wood, 2011) for

modeling the inherent call graph structure.

2.1 Telecom Cloud

Majority of telecom operators manage a large

computing infrastructure comprising of a diverse set

of hardware, software and applications. They

operate in a high availability, performance and

secure environment. In this paper, we propose using

a private cloud to offload the batch processing and

storage of xDR. This provides an ability to handle

peak workloads and release excess computing

capacity for other applications. As discussed, the

private cloud environment can be setup internally. In

order to cost efficiently scale the graph databases,

we are using Amazon’s Infrastructure as a Service

(IaaS) (Garfinkei, 2007) for public cloud setup. The

graph database can be exposed as Software as a

Service (SaaS) for a variety of applications,

customers and partners. The proposed model tackles

the pre-processing and batch operations in an on-

premise private cloud. It hosts the graph database in

a public cloud for generating campaigns.

A campaign lifecycle comprises of 3 major

phases – identify the subscribers, launch the

campaign and follow-up. Potential subscribers for a

campaign can be identified based on segmentation

performed on behavioural characteristics e.g. high

spend, low usage etc. Triggers associated with

business rules can also be used to identify a target

set e.g. send international rate plan promotion to

subscribers who make an international call.

With the available information, campaign

computing based on graphs allow us to quickly

identify subscribers with a certain interest e.g.

subscribers who are interested in a major sporting

event can be identified by traversing the graph with

a search criteria. In addition to simply identifying

the subscribers, we can also get detailed

information

on the ‘neighbourhood’ of subscribers who fit these

criteria. We can explore if these people are part of a

smaller network etc.

Campaign Computing allows operators to

perform a deeper level of analysis. It is tightly

integrated with social networking and forms the

basis of more dynamic marketing models. Current

campaign management systems can design rules to

identify subscribers who are travelling to a different

city and send a promotion. However, graph based

systems can allow the operator to send a reference of

another subscribers from the subscribers circle of

friends who is already using this promotion.

2.2 Graph based Analytics

Graph analytics is the study and analysis of data that

is formed as a network (graph) of objects and

relationships (Modani et al., 2010). Unlike

traditional analytics that focuses on summarizing,

grouping and filtering data, Graph analytics

addresses traversing and navigating the network.

The benefit of using graph database is to quickly

find relationships, connections and patterns between

objects (Saravanan et al., 2011).

InfiniteGraph (Wood, 2011) is a java API used

for creating and managing graph database which

provides fast and efficient query by traversal

operation across massive and distributed data stores.

It supports breaking down navigational queries into

subtasks and executing them across clusters. The

federated InfiniteGraph architecture (collection of

distributed databases) allows elastic scalability of

both processing and data in the cloud environment.

Other major graph database alternatives are

AllegroGraph, graphDB, FlockDB and Pregel.

Telecom call graphs can be modeled with edges

to denote voice calls, connections to websites,

relationship between users etc. Edge weights can be

used to denote call duration, call cost etc. Nodes

designate subscribers. Subscriber information such

as demographics, home location etc. can be stored at

the node level. Often, campaigns are launched using

aggregated data attributes e.g. usage for past 3

months > 100 minutes. In order to accommodate

these cases, aggregated and derived data can also be

stored at the node level.

3 TELECOM CLOUD

ENGINEERING

In this section, we discuss building a telecom cloud.

We use open source Eucalyptus (Nurmi et al., 2009)

to build the private cloud. Amazon Web services are

used for public cloud setup. The internal private

cloud provides an operator complete control over the

environment and enforces stringent security

requirements. The public cloud provides scalability

and access to enterprise users, subscribers and

partners. The confidential data and its pre-processing

DATA2012-InternationalConferenceonDataTechnologiesandApplications

196

are maintained in the private cloud and not exposed

in the public cloud. Fig. 1 describes the proposed

modeling of triggering campaigns.

3.1 Infrastructure

The telecom cloud infrastructure was designed as a

hybrid cloud with commodity servers and Amazon

EC2 instances. A variety of cloud computing

platforms are available including Eucalyptus,

OpenStack, Opennebula and Nimbus (Peng et al.,

2009). Eucalyptus was chosen to build an on-

premise private cloud environment based on the

cloud types, interfaces, compatibility, and

implementation & deployment requirements.

Eucalyptus helps provide a highly available,

scalable, environment compliant with Amazon EC2

APIs. Since the architecture of eucalyptus is simple,

flexible and modular it allows administrators to

deploy and manage machine images and other users

can launch and access virtual instances of the

machine images. Major components in Eucalyptus

architecture are Cloud Controller, Walrus Controller,

Cluster Controller, Storage Controller and Node

Controller.

The Cloud Controller provides access point into

the entire cloud infrastructure for users and

administrators. It queries the node managers for

information about resources, making high level

scheduling decisions and implements them by

making requests to cluster controllers. Walrus

Controller (WS3) is a persistent put/get storage

service. It provides a mechanism for storing and

accessing machine images. Cluster Controller

gathers the information about and schedules VM

execution on specific node controllers, as well as

manages the virtual instances networking. Storage

Controller (SC) provides persistent block. It allows

creating snapshot volumes. Node Controller (NC)

controls the execution, inspection and termination of

VM instances at the host on top of the hypervisor.

3.2 VM Allocation Modeling

Virtualization is the backbone technology for cloud

computing. Hypervisor provides a layer of

abstraction for the Virtual Machines (VM). A VM

encapsulates the resources like Storage, Network,

Hardware, Operating System and pre-Installed

applications. In private cloud environment users are

provided the functionalities to run and control VM

instances and cloud users can access VMs across

various compute nodes. An individual VM core

image was built using KVM hypervisor bundled

with Java, Graph database, Tomcat web server,

Apache Hadoop. In Euca2ools ram-disk, kernel of a

respective platform (OS) is bundled-uploaded-

registered to the private cloud so that the images

with its respective kernel, ram-disk with specified

storage can be spawned as a VM.

3.3 Modeling Dynamics

The large volumes of CDR’s (eg. 200 GB capacity)

were pre-processed using Map Reduce framework

(Jeffrey and Sanjey, 2004) in the telecom cloud. The

subscriber call graphs were partitioned by location

for manageability and performance evaluation. Each

node is named with unique random label. An

MSMapper program written to do simple socket

communication between cloud and operator server

that is maintained in the operator side to map mobile

numbers with unique random labels. Upon returning

the results of the query, the results can be mapped

back to the particular mobile number to trigger

campaigns.

Figure 1: Telecom cloud modelling.

3.4 Security

In order to ensure data confidentiality, critical data is

maintained in the on-premise private cloud.

Eucalyptus provides ingress filtering through the

concept called security groups which are associated

with individual rules with respect to IP/Network,

Protocol type, destination ports etc.

Graph-basedCampaignAmplificationinTelecomCloud

197

Additional users who need to access the cloud

are placed into respective security groups.

Communication with eucalyptus instances occurs via

command line through secured socket shell

connection using public key cryptography.

In addition to the Eucalyptus security model,

MSMapper engine ensure that the confidential

subscriber details are never shared outside the

private internal cloud. Access to the graph database

in the public cloud is governed through standard

user authentication mechanism. The extracted graph

data from private cloud is securely migrated to the

public cloud for faster processing (scalable

environment) of customer details.

4 GRAPH CAMPAIGN

EXPERIMENTS

Initially, the telecom graph data's are preloaded in

the cloud instance with server cluster with high

memory for processing graph campaign

experiments. The telecom data that are visualized in

graph can be trivially represented in relational

database. Querying relational database involved in

larger joins incorporating higher costs for greater

degree of separation. For example, a simple query is

executed to find immediate friends and FOAF with

specific conditions are represented in SQL as well as

in Graph Database and evaluated. The graph based

experiments for immediate friends compared with

respect to RDBMS.

On average to traverse 4673 FOAF paths it took

0.0184s in graph database comparing with RDBMS

took a time of about 0.468s. When the degree of

separation increased to 3, time taken to traverse

160000 dynamic paths of total 667723 edges in a

graph database around 0.957s while in RDBMS it

took 50.59s. On increasing the degree of separation

and total traversal amount RDBMS started to

perform poorly compared to the graph database. The

graph query performance was better when compared

to SQL queries on the provided cloud infrastructure.

5 CONCLUSIONS

This paper explains the exploitation of NoSQL

graph databases in a cloud context for the provision

of campaigns by telecom operators towards their

mobile clients. It also discusses the software

framework for modeling Cloud Computing

environments and an end-to-end Cloud network

architecture for telecom oriented services. Campaign

Computing in Cloud environment using graph

database provides operators a cost-effective,

scalable, secured, advanced analytics platform to

target the telecom customers. This allows operators

to create dynamic, real-time marketing models. The

proposed framework quickly identifies trends,

isolating a targeted subscriber base and rapidly

launching campaigns.

REFERENCES

Zhang, Cheng, Lu, and Boutaba, R., 2010. Cloud

computing: State-of-the-art and research challenges,

Journal of Internet Serv.Appl. 1: 7-18.

Tom, W., 2009. Hadoop: The Definitive Guide, O’Reilly

Media / Yahoo Press, California, USA, 2nd edition.

Leavitt, N., 2010. Will NoSQL Databases Live Up to

Their Promise? Computer, 43:12–14.

Paul Hofmann and Dan Woods, 2010 Cloud Computing:

The Limits of public clouds for Business Applications,

IEEE Internet Computing, 14:90-93.

Armbrust, M., Fox, A., Griffith, R., Joseph, A., Katz, R.,

Konwinski, A., Lee, G., Patterson, D., Rabkin, A.,

Stoica, I., Zaharia, M., 2010. A view of cloud

computing. Communications of the ACM; 53(4):50–58

Darren Wood, Introduction to InfiniteGraph, The

distributed and scalable graph database, 2011. NoSQL

Now!, San Jose, USA

Garfinkei, S, 2007. An Evaluation of Amazon’s Grid

Computing Services: EC2, S3 and SQS. Tech. Rep.

TR-08-07, Harvard University.

Modani, N, Dey, K, Mukherje, S, and Nanavati, A, 2010.

Discovery and analysis of tightly knit communities in

telecom social networks, PIBM Journal of Research

and Development, 7:1-7.

Saravanan M., Prasad G., Karishma S., and Suganthi D,

2011. Analyzing and Labeling of Telecom

Communities using Structural Properties,

International Journal of Social Network Analysis and

Mining, Springer Netherlands, 1-16.

Nurmi, D., Wolski, R., Grzegorczyk, C,. Graziano, O,.

Soman, S,. Youseff, Lamia,. Zagrodnov, Dmitri.,

2009. The Eucalyptus Open-Source Cloud –

Commputing System, 9

International Symposium on

Cluster Computing and Grid.

Peng, J., Zhang, X., Lei, Zhou., Zhang, Wu., Li, Q., 2009.

Comparison of several cloud computing platforms.

2nd International Symposium on Information Science

and Engineering.

Jeffrey Dean, Sanjay Ghemawat, 2004. MapReduce:

simplified data processing on large clusters, Opearting

Systems Design & Implementation, San Francisco,

CA, p.10-10.

DATA2012-InternationalConferenceonDataTechnologiesandApplications

198