ASIC Packet Routing Bypasses Gateway to Reduce Latency
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current cloud service provider infrastructure faces challenges in handling increased bandwidth and port density requirements for remote VPC access, leading to higher latency, decreased per connection consistency, and high management costs due to the need for extensive routing tables and expensive routers.
Innovation Solution
Implementing an offload controller and a programmable switch ASIC that bypasses the gateway by using a VPC controller and virtual route controller to configure routes directly between customer premises equipment and virtual machine hosts, reducing latency and resource usage through fast-path and slow-path packet forwarding.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If all packets in and out of the overlay network are handled at the gateway, then routing control is centralized, but network latency increases and per connection consistency decreases
Solution Approach 1:
The patent segments the routing function by dividing packet handling into two paths: fast-path for established connections that bypass the gateway, and slow-path for new connections that go through the gateway. This segmentation allows routing control to remain centralized for new connections while enabling low-latency forwarding for established connections.
Solution Approach 2:
The patent introduces connection tracking entities as intermediaries between the gateway and the overlay network. These entities maintain connection state information and enable the fast-path forwarding by matching packets against stored connection states, thereby reducing latency while maintaining routing control.
2Reliability
If virtual routing and forwarding is deployed for each VM host to isolate tenant traffic, then tenant isolation is achieved, but the number of routing tables proliferates and management costs increase
Solution Approach 1:
The patent implements a universal connection tracking entity that handles routing decisions for multiple tenants and VM hosts. Instead of maintaining separate routing tables for each tenant, the connection tracking entity uses a unified data structure to track connection states across all tenants, thereby achieving tenant isolation without proliferating routing tables.
Solution Approach 2:
The patent copies only the necessary connection state information to the fast-path forwarding entities rather than replicating entire routing tables. This selective copying reduces the complexity of routing tables while maintaining the ability to isolate and route traffic for multiple tenants efficiently.
3Adaptability or versatility
If x86 systems are used to deploy network function virtualization for flexibility, then new features can be rolled out quickly, but power consumption increases and performance is limited
Solution Approach 1:
The patent replaces the software-based routing processing on x86 systems with hardware-based connection tracking and fast-path forwarding. By implementing connection tracking entities and fast-path forwarding in hardware, the system achieves lower power consumption and higher performance while maintaining the adaptability to deploy new features through programmable hardware configurations.
4Quantity of substance
If routers with statically allocated memory space are used to handle thousands of connections, then routing capacity is increased, but scalability decreases as the number of tenants grows
Solution Approach 1:
The patent implements dynamic memory allocation for connection tracking states, allowing the system to adapt its resource usage based on the actual number of active connections and tenants. Instead of static memory allocation, the connection tracking entities can dynamically allocate and deallocate memory as connections are established and terminated, enabling the system to scale efficiently as the number of tenants grows.
Data Source
AI summary
The techniques described herein include isolating network traffic between tenants using a programmable switch that routes network traffic directly to the appropriate virtual private cloud (VPC) and/or virtual machine (VM).