Multicast Encapsulation for Reliable Multi-Path HPC Routing
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing multicast traffic propagation in multi-path high-performance computing (HPC) networks faces inefficiencies such as excessive packet generation, flooding, and the need for explicit knowledge of all destinations, leading to bandwidth consumption and performance bottlenecks.
Innovation Solution
The system employs encapsulation and de-encapsulation of packets with routing information, allowing for efficient distribution of multicast traffic by minimizing packet duplication and enabling routing around failed links, while using hash checks to ensure reliable delivery and minimize traffic.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If traditional multicast packet distribution is used in multi-path HPC networks, then all nodes receive packets through flooding, but this causes excessive packet generation and bandwidth consumption
Solution Approach 1:
The network is segmented into source switches and destination switches with intermediate switches. Packets are encapsulated with routing information at the source switch and only forwarded through necessary intermediate switches to destination switches, rather than flooding to all switches. This segmentation reduces unnecessary packet transmission while ensuring reliable delivery to intended recipients.
Solution Approach 2:
Encapsulation information acts as an intermediary mechanism that carries routing directives through the network. The encapsulated packets include path information that guides intermediate switches to forward packets only along necessary paths, preventing unnecessary flooding while maintaining delivery reliability through structured intermediate handling.
2Reliability
If packet flooding is used to ensure multicast delivery, then all nodes receive packets, but this creates performance bottlenecks and excessive packet duplication
Solution Approach 1:
The packet distribution is segmented into targeted paths rather than universal flooding. Source switches identify destination switches and intermediate switches, creating segmented transmission paths that avoid unnecessary network segments. This maintains delivery guarantees to intended recipients while preserving network throughput by excluding unrelated nodes from packet reception.
Solution Approach 2:
Instead of excessive action (flooding to all nodes), the system applies partial action by delivering packets only to necessary destination nodes and their associated switches. The encapsulation information enables precise partial distribution, ensuring delivery reliability to intended recipients without the performance degradation caused by universal flooding.
3Productivity
If explicit knowledge of all destinations is required for multicast, then routing can be optimized, but this increases device complexity and operational difficulty
Solution Approach 1:
Destination information is collected and encoded into encapsulation headers in advance at the source switch before packet transmission. This preliminary action allows intermediate switches to simply forward packets based on pre-computed routing information in the encapsulation, rather than requiring complex real-time routing decisions or pre-configuration of all possible destination paths at each switch.
Solution Approach 2:
The encapsulation information serves as an intermediary that carries all necessary routing knowledge through the network. Instead of requiring each switch to have explicit knowledge of all destinations, the encapsulation header acts as a mediator that encapsulates destination information, enabling simple intermediate forwarding while maintaining efficient packet distribution.
4Reliability
If packets are transmitted to all switches in a multicast group, then delivery is ensured, but this causes unnecessary traffic and latency
Solution Approach 1:
The transmission path is segmented to include only source switches, intermediate switches, and destination switches. Encapsulation information identifies this segmented path, preventing packets from traversing unnecessary network segments. This maintains reliable delivery to intended destinations while reducing latency by excluding unrelated switches and links from the transmission path.
Solution Approach 2:
Unnecessary transmission steps are extracted from the packet path. The encapsulation information enables extraction of only the essential forwarding path through intermediate switches, removing unnecessary hops through unrelated switches. This preserves delivery reliability while reducing the time loss associated with traversing unnecessary network segments.
Data Source
AI summary
Provided are systems, methods, and apparatuses for efficient propagation of multicast traffic in multi-path HPC networks. In one or more examples, the systems, devices, and methods include receiving, from a first switch at a second switch, an encapsulated packet comprising information that includes a route to the second switch; obtaining a source packet from the encapsulated packet based on de-encapsulating the encapsulated packet; and transmitting a first copy of the source packet to a first node of the second switch based on copying the source packet. The systems, methods, and apparatuses include distributing multicast traffic to intended participants with the minimal number of replicated packets. Some examples of the systems, methods and apparatuses provide reliable delivery via support for non-posted packets, collating completion responses from participants into a single completion to the multicast packet originator.