Enterprise-Wide QoS Deployment
SLA
A company might use a Service Level Agreement (SLA) to contract with
their ISP for certain levels of service. This typically provides levels of
throughput, delay, jitter, packet loss, and link availability, along with penalties
for missing the SLA. With Layer 2 links (such as frame relay), the
service provider is not involved in providing QoS through its network. With
Layer 3 links (such as MPLS), the service provider can contract for QoS
SLAs through its network.
Service providers use a set number of classes, and your marking must
conform to their guidelines to use QoS SLAs. When calculating the amount
of delay (or latency), jitter, and packet loss for your SLA, remember to take
into account your internal network performance. For example, voice is best
with an end-to-end delay of 150 ms or less. If the latency in the LAN at each
site is 25 ms, then your latency SLA with the ISP should be no more than
100 ms.
Enterprise QoS
Each block within an enterprise network has its own QoS needs and considerations.
In general, you should:
■ Classify and mark traffic as close to the access edge as possible.
Switches that can accomplish this in hardware are more efficient than
routers that must do it in software.
■ Establish the correct trust boundaries.
■ Prioritize real-time traffic (such as voice and video).
■ Configure the appropriate queues on outbound interfaces.
At the Access switch level:
■ Set the trust boundary appropriately.
■ Classify and mark non-VoIP traffic.
■ Place VoIP traffic in interface priority queue.
■ Set speed and duplex.
■ Can use multiple queues, especially on uplink ports.
At the Distribution switch level:
■ Set the trust boundary appropriately.
■ Place VoIP traffic in interface priority queue.
■ Set speed and duplex.
■ Can use multiple queues, especially on uplink ports.
■ Might use Layer 3 policing and marking.
■ Use WRED in data queues.
At the WAN edge:
■ Determine SLA.
■ Might need to reclassify and remark traffic.
■ Use LLQ for real-time traffic.
■ Use WRED in data queues.
■ Might need to use shaping, compression, or LFI.
Within the service provider’s network:
■ Have a DiffServ-compliant backbone.
■ Use LLQ or modified deficit round robin (MDRR).
■ Plan for adequate capacity.
■ Use WRED in data queues.
The actual configuration done on WAN edge routers depends on whether or
not the router is managed by the provider. If it is managed, then the provider
configures output policies on the customer router and does not need any input
policies on the provider edge router. For traffic bound from the provider network
to the customer network, the provider edge router has the configuration to
enforce SLAs.
If the customer edge router is not managed, then customers must configure
their own QoS policies. The service provider likely also configures their edge
router to enforce contracted SLAs on traffic from the customer. For traffic
bound from the provider network to the customer network, the provider edge
router has the configuration to enforce SLAs. The customer might have other
types of configuration, such as reclassifying and remarking to fit their internal
QoS policies.
CCNP ONT