SON Coordination in a Unified Management Framework

Kostas Tsagkaris, Nikos Koutsouris, Panagiotis Demestichas

Dept. of Digital Systems University of Piraeus

Piraeus, Greece {ktsagk, nkouts, pdemest}@unipi.gr

Richard Combes, Zwi Altman Orange Labs

Issy Les Moulineaux, France {richard.combes, zwi.altman}@orange.com

Abstract— The adoption of autonomic features in network management becomes a necessity to cope with the increasing complexity of the telecommunications ecosystem. The purpose of this paper is to present the concept of Unified Management Framework (UMF) developed in the FP7 UniverSelf project. The core of UMF comprises three functional blocks, namely Governance, Coordination and Knowledge which allow to efficiently operate autonomic mechanisms such as SON functionalities in the radio access network. These are denoted as Network Empowerment Mechanisms (NEMs) that, by means of policies, enforce operator business and operation objectives. The paper focuses on the Coordination core block of UMF, and its instantiation to the problem of SON coordination for managing LTE-Advanced a Heterogeneous Network (HetNet) with relays. Prototype description and results are described so as to illustrate the value and applicability of the proposed framework.

Keywords- Unified Management Framework (UMF), UMF Core, Network Empowerment Mechanisms (NEM), Coordination, SON

I. INTRODUCTION Mobile users’ traffic demand increases day by day at a high

pace putting strain to the current network infrastructures. At the same time operators, in trying to confront the situation, have resorted in highly interconnected operation and management systems, which are growing by their own into a really complex structure. It is commonly agreed that the network will not get along with such unforeseen changes. Management solutions that would assist in managing and controlling networks in an effective and flexible way are of utmost importance for those operators that really aim at keeping postulating considerable shares in this emerging technology and business market.

Autonomics has offered as a viable way out namely incorporating the intelligence necessary to tackle such challenges. Autonomic management is based on the distribution of self-x functions (autonomic control loops) in the network management systems and in the network elements per se. However, although the approach has proven to offer flexibility, performance and cost-effectiveness, it also unveils risks associated with delegating the network health guarantee to self-acting, autonomic loops outside the full control of the operator. More specifically, there is a need for a functionality

that ensures the coordinated and conflict-free interworking of multiple autonomic functions that operate simultaneously in the same or interacting domains and possibly affecting the same parameters and/or KPIs. First results on SON coordination for LTE networks have been reported in [1] which investigates the types of conflicts and the corresponding mechanisms required for a SON coordinator.

Univerself [2] is an FP7 European research project that offers a solution to the above problems. More specifically, Univerself specifies a Unified Management Framework (UMF) in the form of functional blocks and interfaces, that will ensure the trustworthy integration (plug and play), operation and interworking (conflict avoidance and knowledge sharing) of multiple autonomic control loops destined to the management of networks and services within the operator's environment.

Moreover, the most characteristic paradigm of deployable self-x functionality in today's networks is Self-Organizing Networks (SON) [3]. SON functions were introduced by 3GPP Long Term Evolution (LTE) as a means for the mobile Network Operators (NOs) to increase the level of automation in their network and decrease their operation and management costs. The efficient, stable and converging interworking of SON (denoted here as coordination) functions in an LTE network is of high significance and as such, it apprehends a prominent position in the design specifications of UMF [4].

Accordingly, this paper focuses on the coordination functionality of UMF, and its instantiation to the problem of SON coordination for managing an LTE-Advanced HetNet with relays. In particular, the paper is split in two parts: The first part describes a theoretical model of a SON coordination problem and of the solution proposed herewith. The second part focuses on the description of the prototype that was developed to support the validation and evaluation of the theoretical models, as well as on the presentation of indicative numerical results.

The rest of the paper is structured as follows: Section II provides an overview of UMF solution. Sections III and IV provide the formal description of the SON and the

978-1-4673-6337-2/13/$31.00 ©2013 IEEE

coordination models considered in this paper. The developed prototype along with a set of indicative results are given in Section V. Finally, the paper is concluded in Section VI.

II. UNIFIED MANAGEMENT FRAMEWORK As already stated, UMF is a framework that aims at

managing multiple, disparate autonomic loops in a unified manner. Univerself bases its UMF design on a set of real telco problems (use cases). In the sequel, it proceeds with the design of a set of autonomic solutions i.e. a set of self-x, methods/algorithms with the aim to solve those problems. These are denoted as Network Empowerment Mechanisms (NEMs). NEMs are designed and deployed with a specific purpose: an operational problem to be solved, a performance objective to be achieved and a network segment or service infrastructure to be targeted. Thus a NEM can be seen as an autonomic network management component with the following functional grouping: method + objective + context (See fig. 1).

When a NEM is deployed, it has to deal with its environment namely, the operator, the network/service equipments, the legacy management systems and also the other NEMs. The main objective of the UMF is to enable a trustworthy integration (plug and play) and interworking (conflict avoidance and knowledge sharing) of NEMs within that environment. In a real operator environment, the number of deployed NEMs can be huge. If targeting the seamless deployment and trustworthy interworking of these NEMs at large scale, one needs to specify functions/mechanisms/interfaces with the following aims:

• to deploy, drive and track progress of NEMs • to avoid conflicts, ensure stability and performance

when several NEMs are concurrently working • to make NEMs find, formulate and share relevant

information to enable or improve their functioning • to allow NEMs retrieving monitoring data and

enforcing configuration actions at equipment level

This has led to the specification of three UMF core blocks: a) Governance, which aims to give a human operator a policy-based mechanism for controlling the network/NEMs from a high level business point of view, without the need of having a deep technical knowledge of the network; b) Coordination, which is responsible for the proper inter-operation of the NEMs: sequence in triggering of NEMs so as to deliver a service (orchestration), avoidance of conflicts between competing NEMs, stability enforcement of inter-working NEMs and the joint optimization of tightly coupled NEMs; c) Knowledge, which is responsible for the unified management of information and knowledge in the UMF system. It provides optimized collection, aggregation, storage, processing and distribution of information. It also performs inferences so as to maintain and provide knowledge services when and where required by UMF operation. Figure 2 depicts an overall functional view of UMF i.e. UMF core blocks, NEMs and the interfaces among them.

NEM_xe.g. SON NEM

NEM_y

GOVERNANCE KNOWLEDGE

networkelement

adaptor

UMF CORE FBs

Objective e.g. ICIC

Methode.g. GA

+

Contexte.g. LTE networks

+

COORDINATION

ICIC: Inter Cell Interference CoordinationGA: Genetic Algorithms

Figure 1. UMF (Core/NEMs) functional view [4]

SON and SON coordination have played a pivotal role in the derivation of UMF design. Specifically, NEMs were designed and developed so as to encapsulate SON specific methods (algorithms) and to provide the interface to "talk to" UMF core. That is to say, SON functions are being instantiated as UMF-compliant NEMs in order to be able to be governed and coordinated and to exchange information/ knowledge within a UMF enabled system.

The analysis and prototyping of two such SON NEMs are presented in this paper. In addition, the paper places focus on the coordination among these SON NEMs. For this purpose, mechanisms are required to populate the coordination block in the UMF core. Such a mechanism is also proposed here. It should be noted that the proposed solution comes with mathematically proven properties of convergence and stability. The latter facilitates building the operator trust in the autonomic system.

III. SYSTEM MODEL Sections II and IIIA are based on results from [4-6], and

present the system model for the two SON functionalities encapsulated in two UMF compliant NEMs

Consider a LTE-Advanced HetNet with relay stations. In band relays are used, namely stations and backhaul share the same frequency resources which are multiplexed in time. Denote by NR the number of relays, and by xs - the portion of time allocated to the backhaul. The index s, 1≥s refers to the relay stations and 0=s - to the eNB (macrocell in this case). The direct (eNB to mobiles and relays to mobiles) links are

active a portion ⎟⎠⎞⎜

⎝⎛ −∑ =

RN

s sx1

1 of the time. We consider the

downlink with elastic traffic. Users arrive randomly according to a Poisson process of intensity λ, and receive a file of size σ bits with a mean size of [ ] .E ∞<σ Mobiles leave the network upon service completion, and mobility is ignored. It is noted that the SON algorithms require only the load information which is insensitive to the file size statistics. The loads can be simulated or obtained using measurements [4-5]. It is assumed that the relay stations are well situated to have favorable propagation conditions with the macrocell with adequate relaying gain [6].

Denote by As the surface covered by station S, by ( )rRs the station peak rate and by ( )rR srel, the backhaul link s capacity at location r .Following [5], we define a quantity related to the loads of the station and backhaul, sρ and srel,ρ , denoted as sρ and srel,ρ respectively

( )( )

( )∫.,

,,

srel

Asrel

As

s R

drrdr

rRr s

s

λρλρ =⎮⌡

⌠= (1)

From (1) one can calculate the corresponding loads:

[ ] [ ]s

s,rels,relN

s s

ss x

,xR

ρσρρσρ

E

1

E

1

=−

=∑ =′ ′

. (2)

IV. SON AND COORDINATION MODELS Relay stations can improve the system performance by

serving part of the macrocell traffic, typically at cell edge, that consume significant eNB resources. By absorbing cell edge traffic, the relays can alleviate eNB load and reduce cell congestion. Two SON functionalities are used to optimize the system performance: the first adapts the amount of traffic served by the relays, and the second adapts the backhaul capacity to forward the adequate amount of traffic to the relays. The Coordination mechanism is used to enforce stable and converging interworking between the two SONs.

A. SON models The first SON is a Load Balancing (LB) functionality that

adapts the coverage zone of the relay stations by adjusting their pilot powers. The dynamics of the LB-SON is described by the Ordinary Differential Equation (ODE):

( ) ( )[ ]PP sss PP ρρ −=⋅

0 , (3)

where Ps is the pilot power of station s, sρ - its load, 0ρ - the macrocell load and P – the vector of stations’ powers. The discretization of (3) defines the SON algorithm: [ ] [ ] [ ] [ ]( )( )nnnPnP snss ρρε −+=+ 011 , (4)

nε being the learning rate, and is chosen here as a constant small number. The properties of the solution, including convergence in the presence of noisy load measurements are obtained using stochastic approximation theorems [4]. In particularly it is shown that upon convergence, the solution of (3) balances the relay stations and eNB loads, and verifies: sss

sρρ minmax = .

The second SON is a Backhaul Resource Allocation (BRA) functionality. BRA adapts the portion of time xs allocated to a backhaul link in order to balance the relay load with its backhaul link load. To this end we define H as the admissible set for xs, { }RNs ,...,1∈ :

⎪⎭

⎪⎬⎫

⎪⎩

⎪⎨⎧

≤≤≥= ∑=

RN

sss xxxH

110,0: , (5)

and denote by [ ]+⋅ H the projection on H. The BRA-SON is written as follows:

[ ] [ ] [ ] [ ]( )[ ]++=+ Hsnss nxngnxnx ,1 ρδ , (6)

( ) ss

N

sss,rels xxx,g

Rρρρ −⎟

⎟

⎠

⎞

⎜⎜

⎝

⎛−= ∑

=′′

11 . (7)

The rationale for (7) is related to the load balancing condition that can be derived from (2):

∑∑ =′ ′=′ ′ −

=−

=RR N

s sNs s

s

s

srel

xxx1

0

1

,

11

ρρρ (8)

B. Coordination model Similarly to the SON functionalities, coordination can be

performed off-line or on-line. Off-line solutions have time scale of the order of an hour to days. They can be implemented in the Network Management System (NMS) where abundant data is available, and use powerful optimization techniques. On-line approaches operate at time scales of the order of tens of seconds to minutes, and are implemented at the control plane (e.g. in the eNB or MME).

Univerself project considers different SON coordination approaches [4], enforced via the Coordination block to ensure stable and converging interworking:

• Hierarchical mode for operating SONs with different time scales; or parallel (tightly coupled) mode with SONs having the same time scale

• Synchronous or asynchronous mode operation

The multiple-SON operation can be described as an online multi-objective optimization when short computation time is required [8]. When the SON functionalities operate synchronously at the same time scale, classical control theory techniques can be used. In [9], the actions of the SON functionalities are formulated as a linear control system. Stability is enforced by correctly modifying the control matrix, and is achieved when the eigenvalues of this matrix become negative.

Separation in time of different SON functionalities have been addressed using a SON-coordination, including minimal time interval necessary for correct SON operation [13]. In the use case considered here, the two SONs, LB-SON and BRA-SON, operate at different time scales. The time scale of LB-SON is fixed according to operational constraints (i.e. avoiding too frequent handovers). BRA-SON uses bigger time steps to guarantee convergence of the hierarchical system, and is seen as quasi static by the LB-SON.

An equivalent solution for the hierarchical operation consists of choosing the same time periodicity for the two SONs, with constant learning rates: εε =n and ,δδ =n which

are chosen as small numbers verifying .0→εδ This solution is equivalent to choosing εδ = and activating LB-SON and BRA-SON with time periodicity of εT and δT respectively, satisfying .0→εδ TT Convergence in distribution to a local optimum solution is can be shown. We refer to this solution as synchronous hierarchical mode of operation.

Using information provided by Governance block, the Coordination block activates the different SON functionalities in one of the coordination mode listed above. In the present use case, the coordination block activates the two SONs using the synchronous hierarchical mode of operation which is defined by the parameters εδ , and the time periodicity T. Convergence properties of the solution in the presence of noisy measurement data can be obtained using stochastic approximation theorems [5].

V. NUMERICAL RESULTS To prove the merits of the proposed solution and to

validate its efficiency, a prototype UMF system has been developed. More specifically, four UMF entities are involved in the experiments through which the numerical results of this section are aggregated: the Governance and the Coordination UMF core blocks, as well as two SON NEMs, the LB_NEM and the BRA_NEM that incorporate the LB-SON and BRA-SON functionalities, respectivelly. The imlementation of the system has followed the second release of the UMF specifications [4] and it is mainly based on Java. It is noted that the two SON functionalities have been developed in Matlab, whereas the integration with the rest of the system, was feasible based on a two-way seamless interaction between a Matlab script and a Java class.

The NO is constantly being informed about the NEMs that have registered with the system, their status and the status of the underlying infrastructure that they manage, through a powerfull Human to Network (H2N) graphical user interface (GUI), one screenshot of which is depicted in Figure 2. This H2N tool includes functionalities that are able to visualize the problems that appear and the actions taken by the various UMF entities, as well as to guide

Figure 2. Screenshot of the UMF H2N GUI

the system’s administrator when a human decision is necessary to be made. For instance, in the case of selecting values for the δ and ε update parameters of the two SONs mentioned in section III, the H2N tool ensures that their ratio remains small.

The setup of the system is carried out on two machines, one for hosting the UMF core and another one for hosting the Matlab environment and the two SON NEMs. Both machines are connected to the same local area network. After the UMF system is initialized, the LB_NEM and the BRA_NEM are executed. During start-up, each NEM sends to the UMF core all the necessary information about its role and capabilities and provides details on the configuration options that can be set in order to control its operation. Then the Governance and the Coordination core blocks can send one or more control policies containing selected values that will ensure the desired behaviour.

The interaction between the UMF core and the UMF NEMs is based on a RESTful API, namely several simple web services that have been implemented using the HTTP protocol and the principles of REST [10]. Therfore, every UMF entity incorporates a light web server called SIMPLE [11] and a suitable client based on RESTY [12]. This choice for implementation was made in order to facilitate the communication of all kinds of devices and equipment with the UMF system.

The underlying wireless infrastructure of the LTE-Advanced HetNet comprises one base station and four relay stations characterized by the parameters given in Table I. Through the H2N GUI the NO is able to activate and deactive the operation of the two NEMs, as well as the functionality related to their coordination. In this way, it is possible to showcase the necessity of coordination and to assess the efficiency of the proposed solution in a qualitative way.

TABLE I. SYSTEM PARAMETERS

The simulator runs for 30 000 time steps (representing

seconds). For each 10 seconds, the SON NEMs are activated. The update parameters, δ and ε are set to 0.001 and 0.03 respectively. They are provided by Governance to the Coordination block to define the coordination mode and activate the SON NEMs in a synchronous hierarchical mode. Without SON, the macrocell has the highest load among all stations (macro / relays). Figure 3. presents the time evolution of the worst link usage (which is equivalent to the load): the

eNB load and the worst backhaul link load with and without SON. Without the enablement of the SON NEMs, one can see that the eNB is highly congested (dashed red line in the upper part of the image), while the backhaul has low load (continuous black line in the lower part of the image). It is recalled that the station with the worst load determines the cell capacity. When the two SON NEMs are activated (see Figure 3. in the middle), the worst direct and backhaul links are balanced, and the maximum link usage is reduced to slightly below 60 percent. As a result, saturation of the macrocell is delayed, giving room to more traffic in the cell. The coordinated SON NEMs successfully balance all the system loads by adapting both backhaul resources and relays’ coverage, which on the average are increased.

100 200 300 400 500 600 700 800 900 10000

20

40

60

80

100

Time (seconds)

Loa

ds (

%)

no SON: backhaulno SON: direct linksSON: backhaulSON: direct links

Figure 3. Link useage for the worst direct (station to mobile) and backhaul links with (two middle curves) and without (top and bottom curves) SON.

We define cell edge users as users with the 10 percent lower throughputs in the cell, and compute the average cell edge throughput. Figure 4. presents the cell edge throughput without and with coordinated SON respectively. Without SON, average cell edge throughut varies from 0.4 to 0.5 Mbps, whereas when activating the synchronous hierarchical operation, above 2 Mbps average cell edge throughput is achieved.

100 200 300 400 500 600 700 800 900 10000

0.5

1

1.5

2

2.5

3

Time (seconds)

Cel

l edg

e us

ers

thro

ughp

ut (

Mbp

s)

no SONSON

Figure 4. Cell edge thoughput with (dashed line) and without (solid line)

SON.

VI. CONCLUSIONS Operators have considered autonomics for enhancing their

telecommunication ecosystem. Yet, the existence and simultaneous operation of multiple autonomic functions (control loops) in the operator's environment may create problems and instabilities if the latter are not properly managed and coordinated. To this end, Univerself has designed a framework for the unified management of multiple, distributed autonomic functions.

The paper briefly describes UMF. Then it focuses on the coordination functionality offered in UMF and particularly, on its instantiation to the problem of SON coordination in LTE-Advanced HetNets with relays. Two SON functions are considered and their model is described first. In the sequel, the model of a coordination mechanism used to ensure a stable and converging interworking among two SONs is also presented. Finally, the paper presents a prototype UMF system implementation which is used to provide qualitative results that prove the concepts and the merits of the proposed functionality.

REFERENCES [1] L.C. Schmelz at al., "A coordination framework for self-organisation in

LTE networks", IM 2011, May 2011. [2] FP7/ICT project UniverSelf, (ICT-2009-257513), Sep. 2010- Aug. 2013,

Website: http://www.univerself-project.eu, Dec. 2012 [3] 3GPP, TR 36.902, April 2011. [Online].

Available:http://www.3gpp.org/ftp/Specs/html-info/36902.htm [4] UniverSelf Deliverable D2.2, “Unified Management Framework (UMF)

specifications, Release 2,” UniverSelf Deliverable, October 2012. [5] R. Combes, Z. Altman, E. Altman, “Self-organization in wireless

network: a flow-level perspective”, IEEE INFOCOM 2012, March 2012. [6] Richard Combes, Z. Altman and E. Altman, “Self-Organizing Relays:

Dimensioning, Self-Optimization and Learning”, IEEE Transactions on Network Management, TNSM, Issue 99.

[7] R. Combes Z. Altman, E. Altman, “Self-organizing relays in LTE networks: queuing analysis and algorithms”, 7th International Conference on Network and Service Management, CNSM 2011, Paris France, Oct. 2011.

[8] G. Athanasiou, K. Tsagkaris, D. Karvounas, P. Vlacheas, P. Demestichas, “Multi-Objective Traffic Engineering for Future Networks,” Communications Letters, IEEE, vol.16, no.1, pp.101-103, January 2012

[9] R. Combes, Z. Altman and E. Altman, “Coordination of autonomic functionalities in communications networks,” http://arxiv.org/abs/1209.1236.

[10] Roy Thomas Fielding, “Architectural Styles and the Design of Network-based Software Architectures”, Doctoral dissertation, University of California, Irvine, Chapter 5, 2000.

[11] SIMPLE framework web site, http://www.simpleframework.org/ [12] RESTY web site, http://beders.github.com/Resty [13] T. Bandh and LC Schmeltz, “Impact-time concept for SON-function

coordination,” 2nd International Workshop on Self-Organizing Networks (IWSON), in IWSC 2012, Aug. 2012.