Flow-Based Local Egress in Multisite Datacenters

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

In multisite datacenters, traffic tromboning occurs when data traffic traverses the datacenter interconnect link multiple times, causing bottlenecks and inefficiencies due to the redirection of migrated workload traffic to the original affinitized datacenter egress edge.

Innovation Solution

Implementing a stateful flow engine in the hypervisor to modify data packet headers, replacing the gateway address of the second site with the original site's gateway address, ensuring traffic follows the original path and avoiding tromboning by directing data traffic to the local gateway and egress edge of the site where the flow originated.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If data traffic from migrated workloads is redirected to the original affinitized datacenter egress edge, then stateful services like NAT are maintained, but traffic tromboning occurs causing bottlenecks and inefficiencies

Engineering Contradiction:
Improvestateful service maintenanceVSAvoiddata traffic efficiency
Core Design Contradiction:
ReliabilityVSProductivity

Solution Approach 1:

The patent applies preliminary action by pre-establishing flow affinity bindings between data flows and egress edges before workload migration occurs. The system records the original egress edge association for each data flow in advance, enabling direct routing decisions after migration without requiring traffic to return to the original site. This proactive binding establishment resolves the contradiction by maintaining stateful service requirements while preventing tromboning effects.

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

The patent introduces an intermediary mechanism in the form of a flow affinity binding table that mediates between migrated workloads and original egress edges. This intermediary structure stores forwarding information and enables direct routing without requiring traffic to physically return to the original datacenter site. The binding table acts as a mediator that preserves stateful service associations while eliminating the inefficiencies of tromboning traffic patterns.

Inventive Principle:
Principle #24Intermediary (Mediator)

2Reliability

If data packets traverse the datacenter interconnect link multiple times to reach the original egress edge, then NAT bindings are preserved, but long latencies and bottlenecks occur

Engineering Contradiction:
ImproveNAT binding preservationVSAvoiddata traffic latency
Core Design Contradiction:
ReliabilityVSLoss of time

Solution Approach 1:

The system performs preliminary action by establishing flow affinity bindings before workload migration, recording the relationship between data flows and their original egress edges. This pre-established binding information enables direct routing decisions after migration, eliminating the need for packets to traverse interconnect links multiple times. The preliminary binding establishment preserves NAT bindings while avoiding the time loss associated with tromboning traffic patterns.

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

The patent applies segmentation by separating the control plane function (maintaining flow affinity bindings) from the data plane traffic flow. The binding information is segmented and stored in a dedicated forwarding information structure, allowing traffic to be routed directly based on pre-recorded associations without requiring repeated traversal through the interconnect. This segmentation preserves stateful service requirements while minimizing latency.

Inventive Principle:
Principle #1Segmentation

Data Source

PatentUS11509581B2Flow-based local egress in a multisite datacenter
Publication Date: 2022.11.22 VMWARE INC
  • US11509581B2 patent drawing
  • US11509581B2 patent drawing
  • US11509581B2 patent drawing

AI summary

A method for a hypervisor to implement flow-based local egress in a multisite datacenter is disclosed. The method comprises: determining whether a first data packet of a first data flow has been received. If the first data packet has been received, then the hypervisor determines a MAC address of a first local gateway in a first site of a multisite datacenter that communicated the first data packet, and stores the MAC address of the first local gateway and a 5-tuple for the first data flow. Upon determining that a response for the first data flow has been received, the hypervisor determines whether the response includes the MAC address of the first local gateway. If the response includes a MAC address of another local gateway, then the hypervisor replaces, in the response, the MAC address of another local gateway with the MAC address of the first local gateway.