The capacity of a SIP trunk is normally defined by the number of simultaneous calls supported and the bandwidth provided for the trunk. An enterprise uses the same Erlang calculations traditionally used in a TDM environment to determine the number of simultaneous calls required on a SIP trunk.
Generally service providers offer a tiered service based on capacity. One of the major benefits of a SIP trunk is that as an enterprise's needs expand, the number of simultaneous calls can be readily expanded without changing the physical interconnection, or even without an increase in provisioned bandwidth, provided excess bandwidth is already available.
Bandwidth Adjustments and Consumption
Bandwidth consumption for IP call traffic inbound from the PSTN on a TDM gateway is easily predicted and controlled because the codec assignment is done by the gateway (or by the enterprise call agent such as CUCM). The use of a CUBE can ensure that this capability is maintained when an enterprise adds a SIP trunk to its communications infrastructure.
CAC policies and features are deployed in the enterprise network based on predictable patterns of codec use by calls (that is, typically G.711 for calls within a site on the LAN and G.729A for calls that traverse the WAN between sites). The bandwidth consumption of inbound SIP trunk calls is partly based on the service provider's configuration, but an enterprise can use a CUBE to influence codec selection (also called codec filtering or stripping) or to transcode streams in the codec selections the enterprise prefers to use.
Call Admission Control (CAC)
Gateways connecting to the PSTN through a TDM interface provide an implicit form of CAC in both directions (inbound and outbound) by virtue of the limited number of channels (or timeslots) physically available on the analog, BRI, T1, or E1 interface. No more calls can simultaneously arrive from the PSTN into the enterprise than there are timeslots available on the gateway TDM trunks, providing implicit call admission control.
With a SIP trunk entering your network on a physical GE connection (possibly fiber or OC3 transport within the service provider's network before hand-off to your network), nothing physical limits the number of calls that could enter or exit your network at any one time.
Top-tier service providers exert CAC control in their networks, and how much protection this offers your enterprise network depends on who your service provider is and how well the controls are implemented. But there is virtually no physical limit, and it is strongly recommended that you protect your own network with your own CAC controls at your Border Element (especially if you are considering a SIP trunk offering without an explicit SLA). This protects against occasional unplanned bursts or surges in legitimate traffic and against potential malicious Dos attack traffic. Lack of CAC control could overrun bandwidth on your network and adversely impact network operations.
One general problem with CAC implementations is that many policies are often based on simple call-counting mechanisms (such as the CUCM Locations CAC feature) as opposed to bandwidth-based mechanisms (such as Resource Reservation Protocol [RSVP]). It is therefore important to control not only the number of calls arriving through the SIP trunk, but also the codec assigned to the calls.
In addition to transcoding and codec filtering, a CUBE can support the CAC policy of the enterprise in the following two ways:
- Limiting calls per dial peer (per destination)
- Limiting calls based on memory and CPU
Limiting Calls per Dial-Peer
You can configure the max-conn command on both the inbound and outbound dial-peers of the CUBE to ensure that no more than the configured number of calls connects at one time. Each call, regardless of codec or the direction of the call, counts as one call.
When a call arrives at a dial-peer and the current number of calls in the connected state exceeds the configured amount, the SIP INVITE request is rejected with a 503 result code to indicate that the gateway is out of resources.
Example 7-7 shows how to configure CAC per dial-peer.
Example 7-7. Using Dial-Peer CAC Mechanisms
dial-peer voice 1 voip max-conn 2
Global Call Admission Control
The CUBE can also be configured to monitor calls on a global basis; that is, without regard of which dial-peer the call might be active on. This global CAC control can be done based on:
- A global system count of calls
- A CPU threshold (as a percentage)
- A memory threshold (as a percentage)
- Any combination of the preceding three metrics
The CUBE checks these configurations and metrics before it completes the processing of a SIP INVITE request. If system resources used exceed the configured amount, the CUBE returns a result code in the SIP INVITE request, indicating that the gateway is out of resources.
Example 7-8 shows how to configure global CAC.
Example 7-8. Using Global CAC Mechanisms
call threshold global total-calls low 20 high 24 call threshold global cpu-avg low 68 high 75 call threshold global total-mem low 75 high 80 call threshold interface Ethernet 0/1 int-calls low 5 high 2500 call treatment cause-code no-resource call treatment on
The call threshold global total-calls command controls the total number of calls to be supported on the CUBE. The command tracks the number of calls, rejecting the 25th call and not accepting calls again until the total number of calls falls below 20. The cpu-avg and total-mem options rejects the calls if the CPU or memory of the border element exceed the given thresholds regardless of the actual active call count. The call threshold interface command limits the number of calls over a specific IP interface.
The call treatment cause-code no-resource command correlates (by default) to a SIP 503 Service Unavailable message sent when calls are rejected.
Quality of Service (QoS)
Cisco provides many methods of measuring and ensuring QoS in an enterprise IP network. You should always use these methods internally when designing a UC system, and you should also extend them to the interconnect point when using a SIP trunk to connect to a service provider. Consider several areas of QoS including:
- Traffic marking
- Delay and jitter
- Congestion management
QoS on IP networks depends on the QoS marking on the IP packets. As with codec settings, QoS markings on voice signaling and media IP packets on IP call traffic inbound from the PSTN on a TDM gateway is easily predicted and controlled by the configuration on your gateway. On SIP trunks, the default packet markings are whatever the service provider sets them to and this might not be in line with your enterprise policies.
The CUBE can re-mark all media and signaling packets that enter you network or exit your network to comply with the SP UNI specification. Re-marking can be done on a per-dial-peer basis (that is, per voice call destination) or per interface (either ingress or egress or both).
Example 7-9 shows how to mark packets per dial-peer.
Example 7-9. Marking QoS on a Dial-Peer
dial-peer voice 40800011 voip destination-pattern 408....... session protocol sipv2 session target ipv4 :10.10.1.1 dtmf-relay rtp-nte
ip qos dscp ef media
ip qos dscp cs4 signalingno vad
Delay and Jitter
The telephone industry standard specified in ITU-T G.114 recommends the maximum desired one-way delay be no more than 150 milliseconds. With a round-trip delay of 300 milliseconds or more, users might experience annoying talk-over effects.
When using SIP trunks, you should consider the IP delay of both the enterprise and service provider networks. In some cases, centralized SIP trunk services cannot be effectively deployed because of the resulting increase in latency. A border element device at the customer premises is required to ensure that latency in the service provider network and enterprise network can be independently measured and controlled.
An echo is the audible leak-through of your own voice into your own receive (return) path. The source of echo might be a TDM loop in the call path or acoustic echo that applies to all-IP calls. Acoustic echo can come from improper acoustic insulation on the phone, headset, or speakerphone (all Cisco IP Phones have an acoustic echo canceller) and is common on PC-based softphones.
A border element demarcation point at a customer site can help you determine if a problem with echo is occurring at the customer premises or in the service provider's network.
When using a single connection for both voice and data, you should carefully consider congestion management (for example, queuing techniques such as Low-Latency Queuing [LLQ]) and bandwidth allocation to prevent data traffic from affecting the voice quality of SIP trunk calls.
The end-to-end voice quality experience of your SIP trunk calls depend on congestion management techniques in both your network and in the service provider's delivery network to your premises. A enterprise border element can help you determine in which network jurisdiction a problem lies.
To ensure business class voice quality within the enterprise network and to determine if a service provider is meeting an agreed-upon SLA, your enterprise should monitor some metrics. Each enterprise might choose to monitor different metrics, but an effective method of collecting the metrics independent from the service provider is important.
Table 7-1 describes some of the important metrics you can monitor. These metrics can be gathered by using various features previously discussed in this chapter in the "Statistics" and "Billing" sections. You can use these basic metrics from the network to calculate the more typical voice quality measurements such as Mean Opinion Score (MOS) or Perceptual Evaluation of Speech Quality (PESQ) to quantify with a single number the voice quality attained by the network.
Table 7-1. Voice-Quality Monitoring Attributes
Method to Monitor
Round-trip delay (RTD)
The RTD is the delay for a packet sent from the originating endpoint at the customer location to the terminating endpoint at the service provider and back again.
This metric can be monitored through the RTD metric in Cisco IOS Software; it is provided per call and is also available through IP-SLA probes.
Jitter is a measurement of the change in the delay of one packet to another during a call.
Jitter is measured in the per-call statistics; the maximum jitter detected during the call is recorded.
1 percent or lower
Packet loss is the number of packets lost during any given call, including UDP and TCP packets.
This metric can be monitored by SNMP in Cisco IOS Software; it is provided per call and can also be tracked with IP-SLA probes.
Uptime is the percentage of time that a path is available for the customer to complete a call to the PSTN.
When uptime is measured, planned outages should be accounted for, and it should be measured as the number of unplanned minutes of outages and monitored with trouble tickets.
Answer seizure rate (ASR) or call success rate (CSR)
The ASR can be recorded as the number of calls made divided by the number of calls that complete a voice path. This number varies greatly because of calling numbers that are unassigned or busy. CSR is the percentage of calls successfully completed through a service provider. The CSR rate should be more than 99 percent. The ASR rate is typically approximately 60 percent.
ASR can be measured by a summary of call activity at the end of the month. The specific value of ASR is not as important as whether there are large swings in the ASR from one month to another that might indicate a problem with end-to-end network connectivity.