Multi-Tenant RoCE Layer-3 Encapsulation for Scalable Cloud Fabrics
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing RoCE implementations in public clouds are limited by the lack of scalability and performance of Layer-2 networks, which do not support multi-tenancy and are prone to issues like head-of-line blocking, network livelock, and congestion management inefficiencies.
Innovation Solution
Implementing Layer-3 routing protocols to enable RDMA traffic over a shared network fabric, using VLAN tags and QoS information to segregate and manage traffic, and employing class-based queuing to maintain separation and performance across multiple tenants.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If Layer-2 network is used for RoCE implementation, then traffic isolation and dedicated queues are achieved, but network scalability and performance deteriorate
Solution Approach 1:
The patent segments the network into multiple virtual networks using VLAN tags, where each VLAN represents a separate tenant's RoCE traffic. This allows Layer-2 isolation benefits to be maintained for each tenant while the overall system operates as a scalable Layer-3 network. The segmentation is implemented through ingress switches that classify and tag packets with appropriate VLAN identifiers.
Solution Approach 2:
The patent transitions from a pure Layer-2 network architecture to a Layer-3 network architecture that carries Layer-2 encapsulated packets. This dimensional change enables the network to achieve both Layer-2 traffic isolation properties and Layer-3 scalability properties simultaneously. The encapsulation adds an outer Layer-3 header while preserving the inner Layer-2 RoCE packet structure.
2Reliability
If dedicated VLANs are used for each tenant, then traffic isolation is maintained, but network complexity and resource overhead increase
Solution Approach 1:
The patent implements a universal network fabric that can handle multiple tenants' RoCE traffic simultaneously using a single Layer-3 infrastructure. Instead of requiring separate physical or logical networks for each tenant, the system uses a universal switch fabric with VLAN tagging and QoS classification capabilities that serve all tenants through a common platform, reducing overall network complexity.
Solution Approach 2:
The patent introduces VLAN tags and QoS fields as intermediary mechanisms that enable traffic isolation and management without requiring complex dedicated network configurations. These intermediaries allow the network to maintain simple Layer-3 routing while providing Layer-2 isolation effects through the encapsulation and tagging process.
3Adaptability or versatility
If Layer-3 routing is implemented over shared network fabric, then network scalability improves, but traffic isolation and performance guarantees may deteriorate
Solution Approach 1:
The patent applies local quality by preserving tenant-specific QoS parameters within the encapsulated packets. Each tenant's traffic maintains its own QoS markings and priority levels, ensuring that performance guarantees are maintained locally for each tenant even though the underlying network is a shared Layer-3 fabric. The QoS fields are copied or mapped during the encapsulation process.
4Productivity
If multi-tenancy is supported on shared host machines, then resource utilization improves, but congestion management and performance isolation become more difficult
Solution Approach 1:
The patent changes the network parameter from Layer-2 to Layer-3 operation, which fundamentally alters how congestion is managed. In the Layer-3 RoCEv2 implementation, congestion control is achieved through standard IP routing mechanisms and QoS policies rather than Layer-2 switch-specific mechanisms. This parameter change enables multi-tenancy on shared infrastructure while maintaining manageable congestion control through well-established Layer-3 protocols.
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.