Per packet load balancing

The sender is the WAN edge device that sends PPL traffic. The receiver is the WAN edge device that receives this traffic and may reorder packets if they arrive out of order.

Defines which specific traffic from the service VPNs use PPL, enabling detailed control through network policies. You can configure this using data policies in the "from-service" direction, and any traffic matched in the policy’s match clause becomes eligible for PPL.

PPL measures the delay across all available tunnels and selects the tunnel with the lowest latency as a reference point. It compares other tunnels to this reference and includes only those within a set latency range for traffic distribution. This method is a Cisco proprietary algorithm. It ensures efficient and reliable path selection for your network traffic.

Instead of splitting traffic into individual packets, PPL groups packets into flowlets, which are small sets of packets. You create flowlets either after sending a set number of packets or when you detect natural breaks (packet gaps) in the traffic.

Packet gap is a method used in PPL to detect natural breaks in a flow of traffic. When there is an idle period between bursts of packets that exceeds the maximum latency difference among available tunnels, a new flowlet is started and sent on a different tunnel.

The group of tunnels available for PPL, selected first by “local colors” and then filtered using metrics like latency and loss. A maximum of eight tunnels can be part of the CTG.

Determines how each flowlet is sent across the reasonably available tunnels in the CTG, using round-robin distribution.

Scenario 1: Managing AI traffic between offices and data centers

AI tasks often involve sending large amounts of information from local offices to a central data center for processing. At the same time, users need quick, real-time responses from AI tools. This traffic is unpredictable and comes in sudden bursts. When all this data is forced through a single network path, it can create network congestion, leading to slow AI performance and delayed data uploads.

PPL breaks these large data streams into smaller, manageable pieces and sends them across all available network connections at once. This prevents any single connection from becoming overloaded.

Scenario 2: High-volume workload transfers and backups in enterprise environments

Enterprise environments routinely perform large-scale data transfers for activities such as workload upgrades, Windows operating system or application updates, data center backups, and disaster recovery operations. These scenarios often require moving significant amounts of data between endpoints. Such transfers can easily exceed the capacity of individual WAN circuits, especially when using multiple mid-sized links.

PPL splits large, single flows into multiple flowlets and distributes them across all available WAN circuits. This approach maximizes total available bandwidth, prevents any single link from being overwhelmed, and ensures that even deprioritized or non-critical traffic, like updates and backups, completes efficiently without impacting business-critical applications.