Network Adapter Head-End Replication for Overlay Multi-Destination Packets
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for handling multi-destination packets in overlay networks are CPU-intensive, latency-prone, and face scalability issues due to dependencies on underlay network support for multicasting, which is not always present, leading to inconsistent performance and impact on unicast packet handling.
Innovation Solution
A network adapter maintains maps of multi-destination groups and tunneling endpoints, provides APIs for manipulating these maps, and handles head-end-replication for multi-destination packets, offloading operations from virtual switches to reduce CPU utilization and improve packet send latencies.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If multi-destination packets are handled by virtual switches in software, then flexibility and control are improved, but CPU utilization increases and latency increases
Solution Approach 1:
The patent extracts the multi-destination packet handling function from the virtual switch software and relocates it to the network adapter hardware. The network adapter maintains maps of multi-destination groups and tunneling endpoints, and performs head-end replication independently, removing this CPU-intensive task from the virtual switch processing path.
Solution Approach 2:
The patent introduces an intermediary mechanism where the network adapter acts as a hardware mediator between the virtual switch and the physical network. The adapter maintains group membership maps and performs packet replication before packets reach the virtual switch, reducing the processing burden on the virtual switch while maintaining control flexibility.
2Ease of operation
If multi-destination packets are handled by virtual switches in software, then control flexibility is improved, but packet send latency increases
Solution Approach 1:
The patent replaces the software-based packet processing mechanism with a hardware-based mechanism in the network adapter. By implementing multi-destination group tracking and head-end replication in hardware, the system achieves faster packet processing and reduced latency while maintaining the same control flexibility.
Solution Approach 2:
The network adapter performs preliminary actions by maintaining maps of multi-destination groups and tunneling endpoints in advance. When multi-destination packets arrive, the adapter can immediately replicate and forward them without requiring real-time software processing, thereby reducing packet send latency.
3Adaptability or versatility
If overlay networks depend on underlay network support for multicasting, then multicast functionality is achieved, but scalability and reliability deteriorate due to inconsistent underlay support
Solution Approach 1:
The network adapter creates copies of multi-destination packets for each destination in the group using head-end replication. Instead of relying on underlay network multicast support, the adapter independently replicates packets at the source, ensuring consistent delivery to all group members regardless of underlay network capabilities.
Solution Approach 2:
The patent segments the multicast delivery function from the underlay network and implements it independently in the network adapter hardware. This segmentation allows overlay networks to achieve multicast functionality without depending on underlay network support, improving reliability and performance consistency across different network environments.
4Ease of operation
If multi-destination packet operations are processed at software level, then flexibility is maintained, but processing efficiency and scalability worsen
Solution Approach 1:
The patent replaces software-based packet processing with hardware-based processing in the network adapter. By implementing multi-destination group management and packet replication in hardware, the system achieves significantly higher processing efficiency and scalability while maintaining flexibility through programmable APIs for manipulating group maps.
Data Source
AI summary
In an embodiment, a network adapter obtains tunneling endpoint information for each virtual switch of a plurality of virtual switches of a multi-destination group. The network adapter encapsulates each of a plurality of replicated multi-destination packets corresponding to respective virtual switches of the plurality of virtual switches with a header specific to a respective tunneling protocol identified in the tunneling endpoint information obtained for the respective virtual switch. The network adapter transmits each encapsulated multi-destination packet to a respective receiver hosted on the respective virtual switch corresponding to the encapsulated multi-destination packet, wherein the respective virtual switch determines a destination port for the encapsulated multi-destination packet on the respective virtual switch by identifying the multi-destination group, determined from the encapsulated multi-destination packet, in a port list on the respective virtual switch.