Programmable Switch Fast-Path Routing for VPC Latency
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current cloud service infrastructure faces challenges in handling increased bandwidth and port density requirements for virtual private clouds (VPCs), leading to higher latency, increased management costs, and scalability issues due to the need for extensive routing tables and expensive routers.
Innovation Solution
Implementing an offload controller (OC) and a programmable switch application-specific integrated circuit (ASIC) to bypass the gateway, allowing for direct packet forwarding between customer premises equipment (CPE) and virtual machines (VMs) within VPCs, using a fast-path and slow-path mechanism to manage traffic and reduce reliance on virtual routing and forwarding (VRF).
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Speed
If all packets in and out of the overlay network are handled at the gateway, then network traffic can be routed to the appropriate VM, but network latency increases and per-connection consistency decreases during frequent VM changes
Solution Approach 1:
The patent segments the gateway's packet handling function by introducing a fast path that bypasses the gateway for established connections. The gateway only handles packet classification and initial routing decisions, while the fast path handles actual packet forwarding. This segmentation reduces gateway processing complexity and latency for data plane traffic.
Solution Approach 2:
The patent introduces a fast path as an intermediary component between the gateway and the overlay network. This fast path maintains connection state and forwarding rules locally, acting as a mediator that handles packet forwarding without requiring gateway intervention for each packet, thereby reducing latency and improving per-connection consistency.
2Reliability
If VRF is deployed for each VM host to isolate tenants' traffic, then tenant traffic isolation is achieved, but the number of routing tables proliferates and consumes excessive space and open connections in the underlay network
Solution Approach 1:
The patent merges multiple VRF routing tables into a single unified routing table at the gateway. Instead of maintaining separate routing tables for each tenant's VRF, the gateway maintains one routing table that routes traffic to different fast paths based on tenant identification. This merging dramatically reduces the number of routing tables from thousands to one, while maintaining tenant traffic isolation through the fast path segmentation.
3Productivity
If routers that can handle thousands and/or millions of connections are deployed, then connection handling capacity is sufficient, but each router costs $1 million or more each
Solution Approach 1:
The patent extracts the connection handling function from expensive dedicated routers and relocates it to the fast path components that can be implemented on standard switching infrastructure. By taking out the packet forwarding function from the gateway and placing it in the fast path, the system achieves high connection handling capacity without requiring million-dollar routers for each connection handle.
Solution Approach 2:
The fast path is designed as a universal component that can handle multiple tenants' traffic simultaneously through a single instance. Instead of requiring dedicated router hardware for each tenant or connection, the fast path provides multi-tenant connection handling capability, reducing the overall system cost while maintaining high productivity.
4Adaptability or versatility
If x86 systems are used to deploy NFV and form the overlay network layer, then flexibility is increased and new features can be rolled out quickly, but management costs and power consumption are high and performance is limited
Solution Approach 1:
The patent replaces the software-based NFV implementation on x86 systems with a hardware-based fast path implementation. The fast path uses dedicated hardware resources (such as network processors or ASICs) to perform packet forwarding and connection management functions, substituting the mechanical/software system with a more efficient hardware system that consumes less power while maintaining flexibility through programmable logic.
Data Source
AI summary
The techniques discussed herein include storing a fast-path and a slow-path table in a memory associated with a programmable switch, such as a cache of the programmable switch. An offload controller may control the contents of the fast-path and/or slow-path table and may thereby control behavior of the programmable switch. The programmable may route a received packet to a gateway if the packet generates a hit in the slow-path table. If the received packet generates a hit in the fast-path table, the packet may be forwarded directly to a virtual private cloud (VPC), virtual switch thereof, and/or to a virtual machine (VM).