Virtual Queue Bandwidth Control for Per-Flow Fair Sharing
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for per-flow fair bandwidth sharing in data center networks and Internet core networks are complex, require multiple queues, and are not suitable for modern switch architectures, leading to inefficiencies and unsatisfactory performance.
Innovation Solution
A method and computing device that manage data transmission by using a list of virtual queues, updating queue sizes based on data packet arrivals, and adjusting credit allowances to enforce approximate fair bandwidth sharing, compatible with both local and end-to-end transmission methods.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If classical fair queuing schedulers (DRR, STFQ) are used to control sharing between traffic classes, then bandwidth sharing control is achieved, but the method does not work for 5-tuple flows and requires one physical queue per-flow which is not implemented in networks
Solution Approach 1:
The patent segments the bandwidth sharing control into two independent layers: (1) a simple FIFO queue structure for packet buffering, and (2) a virtual queue layer that tracks per-flow byte counters for fairness control. This segmentation allows the physical queue to remain simple while the virtual queue layer provides precise per-flow bandwidth control without requiring physical per-flow queues.
Solution Approach 2:
The patent introduces virtual queues as an intermediary layer between the simple FIFO physical queues and the fairness control requirement. The virtual queues track per-flow statistics and control bandwidth allocation, acting as a mediator that enables precise fairness control without complicating the physical queue structure. The virtual queue byte counters serve as the intermediary mechanism that decouples fairness control from physical queue management.
2Ease of manufacture
If a small group of FIFO queues per output are used to classify packets by departure time, then implementation is simplified, but the FIFO queues are not dedicated to flows and the rotating strict priority scheduler is complex
Solution Approach 1:
The patent segments the scheduling function into two independent components: (1) simple FIFO queues that only perform basic packet buffering and forwarding, and (2) a separate credit-based scheduling mechanism that handles fairness control. This segmentation eliminates the need for complex rotating strict priority schedulers while maintaining per-flow fairness, as the FIFO queues remain simple and the fairness control is handled by the credit allowance mechanism.
Solution Approach 2:
The patent extracts the fairness control logic from the queue structure itself and places it in a separate credit-based control mechanism. The FIFO queues are stripped down to their basic function of packet buffering, while the per-flow fairness control is extracted and implemented through virtual queue byte counters and credit allowance adjustments. This extraction simplifies the queue implementation while maintaining fairness control capabilities.
3Manufacturing precision
If remote direct memory access (RDMA) is used for lossless data center networks, then fair bandwidth sharing is enforced, but incoming traffic must be paused at flow granularity which reduces throughput
Solution Approach 1:
The patent applies partial action by adjusting credit allowances proportionally based on virtual queue occupancy rather than completely pausing traffic. When a virtual queue exceeds its fair share, the system adjusts the credit allowance to limit further growth, but does not completely stop packet transmission. This partial control mechanism prevents any single flow from monopolizing bandwidth while maintaining continuous throughput and avoiding the performance degradation associated with complete traffic pausing.
Solution Approach 2:
The patent implements a feedback mechanism where the credit allowance for each flow is dynamically adjusted based on the virtual queue byte counter occupancy. The system continuously monitors the virtual queue sizes and adjusts credit allowances accordingly, providing feedback control that prevents any flow from exceeding its fair share while maintaining overall network throughput. This feedback-based approach replaces the open-loop traffic pausing mechanism of RDMA with a closed-loop control system.
4Manufacturing precision
If HPCC algorithm is used for congestion control, then bandwidth sharing is solved, but a complex end-system algorithm is required to calculate data flow rate
Solution Approach 1:
The patent introduces network-side virtual queue byte counters as an intermediary mechanism that provides fairness control information without requiring complex end-system algorithms. The virtual queues track per-flow occupancy and the network controller adjusts credit allowances based on these measurements, eliminating the need for complex end-system rate calculation algorithms while maintaining precise bandwidth sharing control. The intermediary virtual queue layer handles the complexity centrally rather than distributed at end systems.
Solution Approach 2:
The patent enables self-service bandwidth control where the network infrastructure itself (through virtual queues and credit-based scheduling) provides fairness control without requiring complex algorithms at end systems. The end systems simply send packets and the network automatically enforces fair bandwidth sharing through its internal virtual queue and credit allowance mechanisms, eliminating the burden of complex rate calculation algorithms from end systems.
5Manufacturing precision
If virtual fair scheduler with byte counters is used to track per-flow virtual queues, then fair bandwidth sharing is achieved, but all byte counters need to be frequently updated which is prohibitive in current switches
Solution Approach 1:
The patent extracts the frequent update requirement from the packet forwarding path and places it in a separate, less frequent credit adjustment mechanism. The virtual queue byte counters are updated incrementally as packets are forwarded, but the expensive operation of redistributing credits across all flows is performed only periodically when needed, rather than on every packet arrival. This extraction separates the lightweight counter updates from the heavy credit redistribution operation.
Solution Approach 2:
The patent implements periodic credit adjustment rather than continuous updates on every packet arrival. The virtual queue byte counters are maintained continuously, but the credit allowance redistribution is performed periodically based on the accumulated counter values. This periodic action reduces the automation burden by performing the expensive credit update operation only when necessary, rather than requiring high-frequency updates for every packet event.
Data Source
AI summary
A computer-implementation method includes receiving a data packet; identifying a virtual queue from a list of virtual queues to which the data packet pertains; and determining whether the identified virtual queue size exceeds a threshold maximum size. When the first size does not exceed the threshold maximum size, the identified virtual queue is increased based on a size of the data packet and the data packet is forwarded. The method further includes setting a virtual queue from the list of virtual queues as a target queue; determining a service capacity based on an update time interval and increasing a credit allowance based on the service capacity. The target queue is reduced by an amount based on the credit allowance size, and the credit allowance is reduced by the same amount.