White Paper - QoS Bandwidth Management - White Paper
The following example is intended to illustrate the network model and show how resources are accounted for as calls are established. The customer is a national supermarket chain with a corporate IP-Centrex service. The traffic is identified by a VLAN tag in a 'Metro Ethernet' environment. Each store is a different size and, therefore, may have different traffic requirements.
Figure 1 - Multi-Level Session Admission Control Model
In this scenario, the 1460 session controller can limit the total number of in-store VoIP calls by controlling the service-signalling to and from the Service Provider; this is analogous to buying a PABX with a 20 line capability. Each store can also be limited to a maximum number of external calls placed and registered subscribers connected to the service; this is analogous to the number of external lines that the customer has purchased.
Since the access network between the subscriber and the Service Provider’s core network is often a point of congestion, its usage needs be carefully controlled for technical reasons as opposed to commercial ones. By applying limits to the servicing VLAN, as identified by the VLAN tag, the overall usage purchased by the subscribers can also be limited. In this way, a Service Provider can tailor the subscribers’ package as a whole and on a site-by-site basis.
Bandwidth capacities through the 1460 session controller are modelled within a Link Aggregation Group (LAG) - an aggregation of physical links that connect to a single network. LAG Session Admission Control restricts the number of calls and the bandwidths used in order to remain within the physical capacity limitations of the 1460. This is essential in enabling physical resilience within the LAG. For example, the LAG could be made up of three gigabit Ethernet connections, where the capacity of one of these connections is reserved for resilience. In this case, the SAC limits traffic to be equivalent to two gigabit connections (out of the three available).
Thus, the 1460 session controller’s network model has three stages: subscriber site, total subscriber usage and physical network capacities; at each stage, resource utilisation is accounted for independently. Any over use (or potential over use) results in the new call request being gracefully rejected through call signalling protocol.
Session Admission Control (SAC) and SIP Signalling
SIP signalling uses an ‘offer/acceptance’ mechanism to negotiate the codec type to be used for the call. The SIP ‘INVITE’ message contains a prioritised list of codecs that the calling party offers to the called party. The called party responds with the preferred codec choice and the call proceeds using this codec.
However, there is no mechanism in SIP to reject a call after responding to the initial ‘INVITE’. Therefore, the 1460 performs the resource calculations in two phases:
1. When the session controller proxies the initial ‘INVITE’ message, the most bandwidth-intensive codec from the offered list is used for the first SAC calculation. If this SAC calculation indicates that the system cannot accept the new call, it is rejected.
2. If the call can be accepted using this ‘worst case’ codec, when the called party makes its response, the network model is updated with the bandwidth used by the accepted codec.
Media Policing
Once a call has been admitted, the 1460 session controller uses media policing to underpin the SAC calculation and protect the quality by limiting the bandwidth use per call. This feature also reduces service theft by limiting media bandwidths to that authorised in the signalling setup. In a carrier-class session controller, such as the Newport Networks 1460, over 100,000 calls can be policed independently, using sophisticated hardware acceleration.
Figure 2 - Media Policing
|
IPsec and VoIP
