Switch Packet Replication for Multicast Reliability
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing collective operations in distributed applications, such as multicast and Allgather, face scalability issues due to increased time for packet transmission as the number of recipients grows, leading to inefficiencies in data synchronization and reliability across a packet-switched network.
Innovation Solution
Implementing a switch that replicates packets and tracks acknowledgments, allowing it to confirm receipt to the sender and retransmit packets to recipients that do not acknowledge, thereby offloading processing from the server and improving reliability in multicast and Allgather operations.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If senders send multiple copies of the same data to different recipients, then data delivery is achieved, but scalability deteriorates as transmission time increases with the number of recipients
Solution Approach 1:
The patent introduces a switch as an intermediary device that receives a single packet from the sender and automatically replicates it to multiple recipients. This mediator handles the replication task, freeing the sender from sending multiple copies while maintaining reliable delivery. The switch acts as a smart intermediate node that manages the distribution overhead.
Solution Approach 2:
The switch creates multiple copies of the received packet and forwards them to different recipients simultaneously. This copying mechanism at the network layer allows efficient distribution to multiple destinations without requiring the sender to transmit multiple copies, thereby reducing transmission time while ensuring reliable delivery.
2Reliability
If senders send multiple copies of the same data to different recipients, then data delivery is achieved, but device complexity increases due to server processing load
Solution Approach 1:
The patent extracts the packet replication and acknowledgment tracking functions from the sender/server and relocates them to a dedicated switch. This separation of concerns reduces the processing burden and complexity at the server end, while the switch handles the computationally intensive tasks of replication and reliability management.
Solution Approach 2:
The switch serves as an intermediary that offloads processing complexity from the server. By introducing this intermediate device with specialized replication and acknowledgment tracking capabilities, the server's device complexity is reduced while maintaining reliable multicast delivery.
3Stability of the object's composition
If senders send multiple copies of the same data to different recipients, then data synchronization is achieved, but productivity decreases due to increased transmission time
Solution Approach 1:
The switch performs efficient packet copying and simultaneous forwarding to multiple recipients, enabling data synchronization across all group members without requiring sequential transmissions. This parallel distribution mechanism maintains synchronization stability while significantly improving transmission efficiency and productivity.
Solution Approach 2:
The switch prepares and replicates packets in advance, holding them in buffers until all recipients have received and acknowledged them. This preliminary action ensures synchronized delivery to all group members while optimizing the transmission process for improved productivity.
Data Source
AI summary
Examples described herein relate to a switch circuitry that includes circuitry to cause transmission, to multiple destinations, of copies of a packet received from a sender network interface device and circuitry to indicate acknowledgement of packet receipt, from the multiple destinations, to the sender network interface device based on receipt of acknowledgements of packet receipt from the multiple destinations. In some examples, the circuitry is to indicate acknowledgement of packet receipt, from the multiple destinations, to the sender network interface device with a packet index value.


