RoCE Layer-2 to Layer-3 Encapsulation for Multi-Tenant Scalability
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing RDMA over Converged Ethernet (RoCE) protocols face scalability issues due to reliance on Layer-2 networks, which lack key properties of Layer-3 networks, leading to performance bottlenecks and inability to support multi-tenancy in public cloud environments.
Innovation Solution
Implementing techniques that enable Layer-2 traffic to be communicated over a Layer-3 network using Layer-3 protocols, allowing RDMA traffic to be routed across a shared Layer-3 switch fabric while maintaining class-based separation and quality of service, using VLAN tags and QoS data fields to segregate and prioritize traffic.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If RoCE uses dedicated Layer-2 networks with dedicated VLANs, then RDMA communication reliability is improved, but network scalability and multi-tenancy support deteriorate
Solution Approach 1:
The patent segments Layer-2 RDMA traffic into multiple Layer-3 encapsulated packets, each with separate VLAN tags and QoS values. This segmentation allows different tenants' traffic to be separated and routed independently over the shared Layer-3 network, enabling multi-tenancy while maintaining reliable RDMA communication for each tenant.
Solution Approach 2:
The patent introduces Layer-3 encapsulation as an intermediary between Layer-2 RDMA traffic and the shared Layer-3 network. The encapsulation adds VLAN tags and QoS values that act as mediators, allowing the network to identify, separate, and prioritize different tenants' traffic while maintaining the original RDMA communication reliability.
2Reliability
If RoCE uses Layer-2 networks with dedicated VLANs, then traffic isolation is improved, but network performance and scalability deteriorate
Solution Approach 1:
The patent transitions from a two-dimensional Layer-2 network model to a three-dimensional model by adding Layer-3 encapsulation with VLAN tags and QoS values. This additional dimension allows traffic isolation to be maintained while enabling Layer-3 routing protocols to provide better scalability and performance over shared network infrastructure.
3Adaptability or versatility
If RoCE traffic is transmitted over shared Layer-3 network, then network scalability is improved, but traffic separation and QoS maintenance deteriorate
Solution Approach 1:
The patent nests Layer-2 RDMA packets within Layer-3 encapsulated packets, which contain VLAN tags that identify the tenant. This nested structure allows multiple tenants' traffic to be carried over the shared Layer-3 network while maintaining clear separation and identification of each tenant's traffic through the nested VLAN tags.
Solution Approach 2:
The patent changes the parameter representation by mapping Layer-2 VLAN IDs to Layer-3 VLAN tags and preserving QoS values through the encapsulation. This parameter transformation enables traffic separation to be maintained in the Layer-3 domain, allowing scalable network operation while preserving traffic isolation guarantees.
4Adaptability or versatility
If Layer-2 RDMA packets are converted to Layer-3 encapsulated packets, then multi-tenant routing capability is improved, but packet processing complexity increases
Solution Approach 1:
The patent performs preliminary action by adding the Layer-3 encapsulation with VLAN tags and QoS values at the ingress switch when the packet first enters the network. This upfront encapsulation eliminates the need for complex per-packet processing at intermediate switches, as they can simply forward packets based on the pre-configured VLAN tags and QoS values.
Data Source
AI summary
Techniques and apparatus for data networking are described. In one example, a method includes receiving a first Layer-2 Remote Direct Memory Access (RDMA) packet which includes a virtual local area network (VLAN) tag and a quality-of-service (QoS) data field; converting the first Layer-2 RDMA packet to a first Layer-3 encapsulated packet; and forwarding the first Layer-3 encapsulated packet to a switch fabric. In this method, the converting includes adding at least one header to the first Layer-2 RDMA packet, where the at least one header includes: a virtual network identifier that is based on information from the VLAN tag, and a QoS value that is based on information from the QoS data field.