Synchronized Command Execution in Network Fabric Nodes
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In network fabrics, the unpredictable timing of node reconfiguration due to network traffic and latency issues leads to partial configuration states, resulting in data loss or deadlock situations that may necessitate system resets.
Innovation Solution
Implementing techniques for synchronized execution of commands across nodes in a network fabric by determining separate time periods for each node to receive and execute commands, using time stamps to ensure all nodes reconfigure at the same time, and formatting packets as management datagrams (MADs) compliant with Infiniband Architecture specifications to maintain network time synchronization according to IEEE 1588 standards.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If commands are sent to nodes in parallel without synchronization, then command execution speed is improved, but nodes reconfigure at different times causing partial configuration states
Solution Approach 1:
The fabric manager node calculates separate time periods for each node to receive and execute commands in advance, and includes time stamps in the command packets indicating when each node should execute the command. This preliminary calculation and timing specification ensures that all nodes reconfigure simultaneously at the end of their respective time periods, preventing partial configuration states while maintaining parallel command distribution.
2Reliability
If a hold-down period is implemented to ensure in-order delivery, then frame delivery order is improved, but network reconfiguration time increases
Solution Approach 1:
Instead of implementing a uniform hold-down period across all nodes, the fabric manager node calculates separate, customized time periods for each individual node based on its specific characteristics and position in the network. This localized timing approach ensures that each node has sufficient time to receive and process commands in order, while minimizing the overall reconfiguration time by avoiding unnecessary delays for nodes that can execute commands more quickly.
3Adaptability or versatility
If nodes reconfigure at unpredictable times due to network traffic and latency, then network flexibility is maintained, but data loss or deadlock situations occur
Solution Approach 1:
The fabric manager node sends command packets including time stamps to each node, and each node is configured to execute the command at the specified synchronized time. This feedback mechanism ensures that all nodes reconfigure simultaneously regardless of network traffic variations, preventing partial configuration states that could lead to data loss or deadlock while still allowing the network to handle traffic dynamically during the reconfiguration process.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
Examples may include techniques to enable synchronized execution of a command by nodes in a network fabric. A node capable of hosting a fabric manager for the network fabric (fabric manager node) may generate one or more packets including a command to be executed by at least some nodes in the network fabric. In some examples, a time stamp is also included with at least one of the one or more packets to indicate to receiving nodes to execute the command at a synchronized time.