Target Group Congestion Control for HPC Fabric VM Migration
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing technologies face challenges in efficiently managing congestion and ensuring quality of service (QoS) in high performance computing environments, particularly in InfiniBand and RoCE fabrics, which affect the live migration of virtual machines due to complex addressing and routing schemes, leading to performance degradation and connectivity issues.
Innovation Solution
Implementing a system and method for congestion control in private fabrics by defining target groups with bandwidth limits, using a target group repository, and employing dynamic reconfiguration mechanisms to minimize overhead and ensure seamless live migration of virtual machines.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If traditional network addressing and routing schemes are used in high performance computing fabrics, then basic network connectivity is maintained, but performance degradation and administrative bottlenecks occur during live migration of virtual machines
Solution Approach 1:
The patent segments the network addressing scheme by introducing fabric-wide unique identifiers (FWIDs) that are independent of traditional subnet-based addressing. This segmentation allows virtual machine migration across subnets without requiring complex routing reconfiguration, as the FWID remains constant regardless of physical location. The network is divided into logical domains where addressing is decoupled from physical topology, enabling efficient live migration while maintaining basic connectivity through traditional mechanisms.
2Reliability
If bandwidth limits are enforced on inter-switch links to ensure QoS, then quality of service is improved, but network flexibility and resource allocation efficiency deteriorate
Solution Approach 1:
The patent implements dynamic bandwidth allocation through target groups that can be configured with specific bandwidth limits on inter-switch links. These limits are not static but can be adjusted based on current network conditions and service requirements. The system dynamically routes traffic through alternative paths when bandwidth limits are approached, and allows elastic resource allocation where bandwidth can be borrowed from underutilized links. This dynamic approach maintains QoS guarantees while preserving network flexibility and adaptability.
3Measurement precision
If complex addressing schemes are used to support virtual machine migration, then routing precision is improved, but administrative overhead and reconfiguration time increase
Solution Approach 1:
The patent employs preliminary action by pre-allocating fabric-wide unique identifiers to virtual machines before migration occurs. These identifiers are registered in advance in a centralized directory service that all network nodes can query. When migration is initiated, the new physical location is already known to the network infrastructure through pre-established routing tables that map FWIDs to current physical addresses. This eliminates the need for complex real-time address translation and routing recalculation during migration, maintaining precise routing while minimizing reconfiguration time to near-zero.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
Systems and methods for supporting target groups for congestion control in a private fabric in a high performance computing environment. An exemplary method can provide, at one or more microprocessors, a first subnet, the first subnet comprising a plurality of switches, a plurality of host channel adapters, and a plurality of end nodes, including a plurality of virtual machines. The method can define a target group on one of an inter-switch link or at a port of a switch of the plurality of switches, wherein the target group defines a bandwidth limit on the at least one of an inter-switch link between two switches of the plurality of switches or at a port of a switch of the plurality of switches. The method can provide a target group repository stored in a memory of the host channel adapter where the defined target group in the target group repository is recorded.