Interconexión al Sistema

As discussed in section 2.1, aQoSarchitecture with an admission control mech-

anism can be used in order to avoid congestion and thrashing situations in networks with inelastic or elastic multimedia traffic. In wired networks, over- dimensioning is often considered as an alternative in order to minimize the probability of congestion. In today’s networks however, there is an increasing use of wireless technologies. Here, over-dimensioning is not an option due to the lack of frequency resources and limitations on channel capacity. In addition,

in wireless environments such as in mobile ad-hoc networks (MANETs), there

are often high dynamics caused by mobility and interference. Consequently, the channel capacity is varying with time, and the probability of congestion and the need for admission control is substantial. The dynamic environment with variations in channel conditions and likelihood of re-routing events also increases the need for termination of already admitted flows.

As discussed in the related work section (3.1.2), several admission control

solutions forMANETshave been proposed in the literature, targeting the range

ofMANET-specific challenges not found in wired networks. There are however a

number of reasons why Pre-Congestion Notification (PCN) (described in section

2.1.2) should be considered as an alternative to those solutions. Recall thatPCN, and also DiffServ, are primarily designed for wired networks.

First of all, PCN is likely to be used in the wired networks to which the

MANET is directly connected, as illustrated in figure 3.1. When using the

same standardizedQoS framework in all parts of the network, both capital and

operational costs may be reduced if assuming that both the wired and the wireless networks belong to the same service provider. Less effort is needed in order

to map between two differentQoStechnologies, and less resources are spent on

education of the operators.

A second advantage of usingPCNis that it comes with a built-in flow termina- tion mechanism, in contrast to a large part of its alternatives. InMANETs, where significant reduction in capacity is likely due to shadowing effects, mobility, re- routing and background noise, the conditions during which an admission decision

3.1. Network level congestion control

A B C

D

New flow requested

E

F G

Figure 3.2. The hidden path issue. The curves mark the sensing range of node B and E respectively. Node D is in a critical area because it is impacted by both traffic admitted on the path from A to C and traffic on the path between F and G.

was taken may change drastically during the admitted flow’s lifetime. The ability to intelligently select admitted flows for termination is therefore important.

Another property ofPCNmotivating its use inMANETsis the independence

from link layer and routing protocols. This is important in a standardization context.

There are numerous challenges with distributed admission control inMANETs.

The most critical ones are listed below:

• Channel variations. There are multiple reasons why the capacity on a link

between nodes int aMANETis time varying. The mobility causes changes in

the path-loss as well as changes in the shadowing effects caused by obstacles. In addition, there may be changes in external noise around the receiver. In turn, these channel variations may cause link breaks and re-routing of admitted flows onto new paths. Thus, in addition to the varying capacity, there is also a large degree of varying load. Consequently, over-admission may occur on a link even if no new flows where accepted into the network. There are two alternative ways to deal with this issue. The first is to have a conservative admission limit. This will result in good performance for the admitted flows, but high rejection rates and low utilization of the network resources. The other alternative is to rely on a flow termination mechanism to remove flows when congestion occurs. This will lower the rejection rates and increase the utilization of the network resources, but lower the performance of the admitted flows.

links, but in wireless networks, the channel is shared with all the neighbor nodes within the sensing range. It then becomes a collective responsibility to ensure that the capacity is not exceeded on the shared channel around any node. When assessing whether a node can admit a new flow, it not only needs to evaluate the resource usage on the channel around itself, but also around all of its one hop neighbors. This is valid for all nodes along the path of the new flow. Consider the example in figure 3.2. A new flow is requested from A to C through B. The channel load as observed by node B is low, and there is room for the new flow. However, the new flow will also affect node D, which is a 1-hop neighbor of B and is located in a critical area; The load around D may be higher due to traffic on the hidden paths through node E. Consequently, for the correct decision to be taken, it is necessary to take into account the traffic load around nodes in the critical area. Probing or other signaling along the requesting flow’s path only is not sufficient, as extra signaling between neighbors is needed.

• Empty ingress-egress aggregates (IEAs). In wireless networks, where flow

sizes are typically large relative to the network capacity, only a small number of flows may be admitted in the system compared to fixed high capacity networks. This means the likelihood of having source-destination

paths (“IEA” isPCNterminology) without any admitted flows is significant.

As many admission control schemes (includingPCN) is based on monitoring

existing traffic flows in order to collect information about congestion status, the basis of decisions is missing when paths have no traffic. Often the solution is to simply admit new flows in these situations, which is likely to cause over-admission.

• No distinct core/border separation. In fixed DiffServ networks, scalability

issues are avoided by having a fast and simple core, while the edge routers perform more complex per-flow tasks such as classification, admission

control, policing etc. Distributed admission control schemes (e.g. PCN)

normally involves signaling between the border routers. In aMANET, the

distinction between edge and core nodes is not present in the same way. Instead, the nodes take different roles with respect to each flow. E.g, a node may operate as an ingress router for one flow, while having the role as an intermediate node for another flow. Consequently, all nodes are potential edge nodes. This impacts scalability for distributed admission control schemes since all node pair must exchange signaling.

The work in paper A is motivated by need for admission control solutions for MANETs dealing with these challenges, combined with the identified advantages

of using the PCNmentioned above. The paper reviews the possibility of using

PCNinMANETs, as well as suggesting improvements and adjustments in order to enhance the performance.

3.1. Network level congestion control