Flexible P_Key Mapping for InfiniBand Scalability
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The limitations of traditional InfiniBand networks in supporting large-scale cloud computing environments, particularly due to the 48K unicast LID value space constraint, which hinders the scalability and flexibility in routing and load-balancing across multiple subnets, leading to performance bottlenecks and administrative challenges.
Innovation Solution
The implementation of a system and method for flexible partition key (P_Key) mapping within a high-performance computing environment, utilizing a global fabric manager to define inter-subnet partitions and update packet P_Key values for efficient routing across subnets, enabling scalable and flexible routing and load-balancing through dynamic reconfiguration of network paths.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Quantity of substance
If hierarchical mapping structure is used to scale to large number of destinations, then scalability of subnet size is improved, but ability to route individual destinations independently deteriorates
Solution Approach 1:
The patent segments the routing function into two independent tables: a first routing table for hierarchical subnet mapping (scaling to large destinations) and a second routing table for individual destination routing (maintaining routing flexibility). This segmentation allows both scalability and independent routing capability to coexist by dividing the routing decision process into two stages.
Solution Approach 2:
The patent introduces an intermediary routing mechanism where the first routing table acts as a mediator that maps hierarchical subnet information to enable lookup in the second routing table. This intermediary layer allows the system to benefit from both hierarchical scaling and fine-grained routing control without direct conflict between the two functions.
2Speed
If content addressable memory based lookup hardware is used for DGID to DLID mapping, then routing speed is improved, but hardware complexity and cost increase
Solution Approach 1:
The patent replaces the mechanical/CAM-based lookup hardware with a software-based routing table lookup mechanism. Instead of using complex content addressable memory hardware for DGID to DLID mapping, the system uses programmable routing tables that can be updated dynamically, substituting hardware complexity with software flexibility while maintaining acceptable routing performance.
Solution Approach 2:
The patent changes the dimension of the routing solution from hardware-based parallel lookup (CAM) to a two-stage table-based approach that uses sequential processing with optimized data structures. This dimensional shift from hardware to software architecture reduces hardware complexity while maintaining routing functionality.
3Stability of the object's composition
If fixed partition key mapping is used in multi-subnet fabric, then routing stability is improved, but flexibility in load-balancing and QOS deteriorates
Solution Approach 1:
The patent introduces dynamic partition key mapping where the second routing table can be updated with multiple P_Key values for the same destination LID based on current network conditions, traffic patterns, and QOS requirements. This dynamic updating capability allows the system to maintain routing stability through consistent destination mapping while achieving load-balancing flexibility by selecting different P_Key values dynamically.
Solution Approach 2:
The patent changes the parameter of partition key mapping from fixed to variable, allowing multiple P_Key values to be associated with a single destination LID in the second routing table. This parameter change enables the system to adjust routing behavior dynamically while maintaining stable destination reachability, thus achieving both routing stability and load-balancing flexibility.
Data Source
AI summary
System and method for supporting flexible partition key (P_Key) mapping in a high performance computing environment. In accordance with an embodiment, when such computer environments comprise more than one subnet and/or forwarding domain that support inter-subnet partitions, the P_Keys comprising such inter-subnet partitions can change across domain boundaries. In such situations, a partition key of a packet can be updated upon entry to a new subnet and/or forwarding domain.