To compare the different solutions, different key performance indicators are measured, such as the transport handover delay, the mean bandwidth and the volume of data downloaded during the simulation time (500s). Only results regarding the downloaded data volume indicator are given, as this indicator is the most global and pertinent one contrary to the transport handover delay. Indeed, the transport handover delay impacts the volume of downloaded data but is not enough to reflect it, as this latter depends on cwnd evolution which may remain constant after handover instead of increasing exponentially (see section 7.6.2).
Firstly, m-SCTP, m-SCTP+, SxS, SxS+ and SxS++ are compared for the network scenario Sc1 (see table 7.1), considering different values for the handover number (HO_nbr). A HO_nbr value corresponds to a given MN velocity. For example, HO_nbr=6 corresponds to a pedestrian walking at 5km/h in an area covered by 100m-diameter cells or a user in a car traveling at 51km/h in 1km-diameter cells.
Figure 7.7 gives the additional data volume m-SCTP+, UFA, SxS+, SxS++ enable to download compared to m-SCTP. For m-SCTP, the downloaded data volume is 62, 61, 60, 59, 57 Mbytes for HO_nbr equal to 1, 3, 6, 9, 13 respectively.
It can be observed that all UFA configuration options enable to download more data than m-SCTP. Moreover they are more efficient than m-SCTP+, considered in the state of the art as the best improvement to m-SCTP performance. Indeed, m-SCTP+ enables a gain ranging from 0.2% to 2% compared to m-SCTP, whereas UFA enables a gain ranging from 0.4% to 7.8% compared to m-SCTP.
For this network scenario Sc1, SxS+ and SxS++ do not provide remarkable gains compared to
SxS as both DGW −CN and BDP are low, which enable packets lost in SxS to be recovered rapidly
(SACKs are returned rapidly) and cwnd to attain rapidly BDP.
Therefore, the performances of SxS, SxS+ and SxS++ are compared using the network scenario Sc2
(see table 7.1) that considers higher values for DGW −CN and BDP (higher value for XputGW −MN).
The receiver buffer size in the MN is set to 200000bytes, in order to take into account the high
bandwidths (XputGW −MN equal to 2Mbps or 3Mbps). Figure 7.8 shows the additional data volume
7.6. PERFORMANCE EVALUATION 135 A C E F m-SCTP UFA SxS B D UFA SxS m-SCTP UFA SxS m-SCTP
Figure 7.5: TSN on the CN side for m-SCTP and UFA SxS configuration option in the network scenario Sc1
Figure 7.6: Cwnd for the different UFA SCTP configuration options in the network scenario Sc2
136
CHAPTER 7. PERFORMANCE OF UFA MOBILITY PROCEDURE FOR NON-SIP NATIVE SERVICES
Figure 7.7: Comparison of m-SCTP, m-SCTP+, SxS, SxS+, SxS++ in the network scenario Sc1
7.7. CONCLUSION 137
6, 28, 57, 116, 168 Mbytes for XputGW −MN equal to 0.1, 0.5, 1, 2, 3 Mbps respectively. It can be
observed that, compared to SxS:
• SxS+ enables a gain varying from 2% to 7% for XputGW −MN varying from 1 to 3Mbps, and
• SxS++ enables a gain varying from 4% to 9% for XputGW −MN varying from 1 to 3Mbps.
Thus, SxS+ and SxS++ provide better performances than SxS. Moreover, the benefits of SxS+ compared to SxS are more important than those of SxS++ compared to SxS, given the fact that SxS+ enables to skip the important time period during which cwnd is constant (see figure 7.6). In general, although the additional downloaded data volume may appear relatively low (1...14Mbps), this one shall not be neglected as it has been calculated for a short downloading time period (simulation time=500s) and for a single MN. The gain for an operator is important as it is proportional to the number of MNs and to the duration of data downloading (higher than 500s).
7.7
Conclusion
This chapter has underlined the strengths of UFA compared to m-SCTP and other SIP-based solutions managing the mobility of SCTP-transported services and optimizing their performances during handover. UFA relies on SIP, a network-controlled mobility procedure and cross-layer mechanisms to optimally drive SCTP configuration on the MN and the CN during handover. Three incremental UFA SCTP configuration options are conceived. The first option, shortly named SxS, is the basic one. It configures SCTP layer with the new MN IP address and reduces the handover delay. In SxS+ based on SxS, SCTP on the CN side is enhanced by immediately sending lost packets through the new link, before new packets. SxS++ based on SxS+, additionally updates SCTP congestion control parameters according to the new link bandwidth.
SxS++ solves simultaneously the problems encountered by SCTP-transported services when used with m-SCTP, that are packet losses due to handover delays and bandwidth under-utilization due to link change after handover.
All UFA SCTP configuration options have been implemented and compared to m-SCTP and its best improvement (m-SCTP+), proposed in the state of the art to deal with the problems due to handover delays. Performance results are promising:
(1) All UFA SCTP configuration options provide better performances than m-SCTP. (2) SxS is even better than enhanced m-SCTP solutions (m-SCTP+).
(3) Both SxS+ and SxS++ improve UFA performance.
The gain obtained with UFA configuration options is important for an operator; it is proportional to the number of MNs and the duration of data downloading.
UFA principles can be also applied for TCP-transported services for configuring their congestion control parameters, but in this case a solution has to be defined to hide the IP address change from TCP.
Part IV
CHAPTER
8
Conclusion and future
work
8.1
Thesis summary
This thesis has highlighted the importance of flat network models in enhancing the readiness of mo- bile network operators for the new telecommunication ecosystem, characterized by an exponential traffic increase. Different steps, given below, have been necessary to reach such a conclusion.
Specifying requirements for mobile networks to face the ecosystem chal-
lenges
• Characterizing the new telecommunication ecosystem: the work in this thesis has started with characterizing the new telecommunication ecosystem in which mobile networks evolve. It has also raised the need to analyze the mobile networks readiness to face the challenges of this ecosystem. The first challenge is the exponential traffic increase due to the high bitrate radio interfaces, the emergence of high bitrate applications and the increasing number of mobile users. The second challenge is the stagnation of operator revenues due to the adoption of new business models based on flat rates. Part of these challenges has been the object of [38, 80].
• Modeling current mobile networks and analyzing their readiness for the new
ecosystem: mobile networks have been modeled in order to analyze their readiness for the
new ecosystem. Mobile networks model (IP-AN//PCC//IMS) includes the IP access network (IP-AN) offering IP connectivity to users, the IMS providing service control for the IP-ANs, and the PCC ensuring the interaction between the IP-AN and the IMS with regard to the policy control function.
Current IP access networks are characterized by a centralized anchor node (first IP router) and different intermediate anchor nodes implementing specific functions and hiding the users mobility from the higher level anchor nodes. IMS and PCC nodes are inevitably centralized, as they are located in the IP network behind the IP access network.
Four criteria have been considered in analyzing this IP-AN//PCC//IMS network model: ser- vice control, scalability, service and network convergence, QoS. It has been observed that, to meet these criteria, this model needs to implement additional functions at the expense of more complexity and a non-idealistic solution.
The model is not scalable as it is centralized and contains numerous node types and interfaces. Moreover, it lacks a mobility procedure between the centralized anchors (first IP routers),
142 CHAPTER 8. CONCLUSION AND FUTURE WORK
which are initially designed to be unchanged whatever the user mobility is (e.g. GGSN in UMTS).
The model suffers from the following QoS problems: a long delay to access services, a high handover delay between the centralized nodes (e.g. GGSN), and a service non-adaptation to the resources available in the IP Access network. These problems are due to a high number of signalling messages involved in the service access and mobility procedures, the fact that these messages cross the IP-AN//PCC//IMS centralized nodes, and the fact that the service control layer (responsible for service adaptation) doesn’t interact well with the resource information located in the different IP access network nodes.
Part of this analysis has been published in [38, 80, 81, 85].
• Specifying requirements for mobile networks to face the ecosystem challenges: The analysis of how the IP-AN//PCC//IMS model meets the defined criteria (service control, scalability, QoS, service and network convergence), has led to setting respectively different requirements. These have to be fulfilled by a mobile network in order to face the ecosystem challenges better:
– Requirement 1: the service control shall be provided to all applications in a cost effective
manner. IMS is a service control solution already available for SIP native applications. It is proposed to extend it to all applications.
– Requirement 2: the mobile network has to be scalable. It means that, in case of huge
data growth, network investments shall remain profitable for operators.
– Requirement 3: a simple and optimal mobility procedure between the first IP routers,
handling the same IP-AN type (e.g. between GGSNs in UMTS) or different IP-AN types has to be supported.
– Requirement 4: service and network convergence shall be provided.
– Requirement 5: the service access procedure shall be optimized, in terms of necessary
signalling messages, tasks and mapping functions.
– Requirement 6: the number of nodes within the mobile network model shall be reduced
in order to enhance the service access and the handover delay.
– Requirement 7: centralized nodes shall be avoided i.e. nodes shall be rather distributed
in order to enhance the access and the handover delays.
– Requirement 8: a tight interaction between the service control layer (IMS) and the net-
work layer (IP-AN) shall be possible, without making the network more complex. This would allow a reactive service adaptation solution.
– Requirement 9: load information as well as other handover decision inputs shall be made
easily available to the handover decision function.
Part of these requirements have been stated in [38, 80, 81, 85].
Proposing a new model (UFA, Ultra Flat Architecture) for future mobile
networks
The above requirements have incited me to review the IP-AN//PCC//IMS network model and to define a new model, called UFA (Ultra Flat Architecture).
UFA is a flat model that uses IMS as a unified service control layer for all applications. Therefore, it is entirely based on SIP. In UFA, the classical IP-AN node functions (e.g. NodeB, RNC, SGSN