Explains how per-class application-aware routing uses SLA metrics and policies to direct application traffic based on loss, latency, and jitter measurements for accurate network performance.
A per-class application-aware routing policy is a network routing mechanism that
-
uses SLA metrics such as loss, latency, and jitter measured between TLOCs
-
applies policies that constrain paths for forwarding applications based on SLA classes, and
-
requires active probing or passive monitoring to measure metrics on all paths to the traffic destination.
SLA metrics overview
The SLA definition includes loss, latency, and jitter values measured using the BFD channel between two TLOCs. These values collectively represent the status of the network and the BFD link. The system sends BFD control messages with a high-priority Differentiated Services Code Point (DSCP) marking of 48.
However, SLA metrics derived from these high-priority packets do not reflect the priority that actual data traffic receives as it flows through the edge device. Depending on the application class, data traffic can carry different DSCP values in the network. Therefore, the network requires a more accurate representation of loss, latency, and jitter for real traffic profiles so it can direct different traffic types to the appropriate tunnels.
Application-aware routing and SLA policies
Application-aware routing uses policies to constrain the paths available for forwarding applications. These constraints are typically defined as SLA classes containing loss, latency, and jitter requirements that the network must meet.
To enforce these policies, the system measures SLA metrics across all paths to the traffic destination. It performs these measurements using either active probing or passive monitoring.
Active probing methods
Active probing generates synthetic traffic and injects it alongside real traffic. The network expects both probe traffic and real traffic to follow the same forwarding behavior.
Examples of active probing mechanisms include:
-
BFD probing
-
ICMP probes
-
Periodic HTTP requests
-
IP SLA measurements
The Cisco Catalyst SD-WAN solution uses BFD-based probes for active measurements.
Passive monitoring methods
Passive monitoring relies on the Cisco Catalyst SD-WAN Application Intelligence Engine (SAIE) to observe real traffic flows and measure loss, latency, and jitter directly from live traffic. For example, the system monitors RTP and TCP traffic to gather these metrics.
In Cisco vManage Release 20.7.1 earlier releases, the SAIE flow was referred to as the deep packet inspection (DPI) flow.