Network Adapter Common Queue for RDMA and Processing
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In high-performance computing networks, existing technologies face inefficiencies and latency issues when handling combined data processing and remote communication work requests, as they require the host processor to manage separate queues, leading to bottlenecks and increased processing cycles.
Innovation Solution
A network adapter with processing circuitry that executes data processing and RDMA work requests in a common work queue, allowing concurrent queuing of different types of work requests, thereby offloading processing tasks from the host processor and optimizing data transfer operations.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If separate queues are used for data processing and RDMA work requests, then work requests can be organized by type, but the host processor becomes a bottleneck and latency increases
Solution Approach 1:
The patent combines separate data processing work queues and RDMA work queues into a single unified work queue. This merging eliminates the need for the host processor to separately manage and switch between multiple queues, reducing processing overhead and latency while maintaining reliable handling of both work request types through a single streamlined queue structure.
Solution Approach 2:
The unified work queue is designed to handle multiple types of work requests (data processing and RDMA operations) simultaneously. This universal queue structure allows the network adapter to process diverse work items through a common interface, eliminating the bottleneck of host processor-mediated queue management while preserving the reliability of individual operation types.
2Ease of operation
If the host processor manages separate queues for data processing and RDMA operations, then work requests can be properly routed, but processing cycles increase and performance decreases
Solution Approach 1:
The patent extracts the queue management function from the host processor and relocates it to the network adapter's local queue structure. This extraction eliminates the host processor's involvement in queue management overhead, allowing work requests to be routed and processed directly by the network adapter, thereby improving processing throughput while maintaining proper routing through the unified queue.
Solution Approach 2:
The network adapter performs self-service by autonomously managing the unified work queue without requiring host processor intervention. The adapter independently handles work request routing, processing, and execution, which eliminates the performance bottleneck caused by host processor involvement while ensuring correct routing through the queue's inherent structure.
3Adaptability or versatility
If multiple separate queues are maintained, then work request organization is improved, but device complexity and memory usage increase
Solution Approach 1:
The patent merges multiple separate work queues into a single unified queue structure, reducing device complexity and memory usage. The unified queue maintains the ability to organize and handle different work request types through a single streamlined structure, eliminating the complexity of managing multiple separate queues while preserving adaptability through the queue's versatile handling capabilities.
4Reliability
If the host processor is involved in every work request processing, then proper control and management are ensured, but bandwidth utilization and performance are reduced
Solution Approach 1:
The patent extracts the work request processing function from the host processor and transfers it to the network adapter. This extraction ensures that the host processor is involved only in high-level control decisions, while the network adapter independently handles the actual processing of data processing and RDMA operations, thereby improving bandwidth utilization and overall performance while maintaining proper control through the adapter's autonomous operation.
Solution Approach 2:
The network adapter performs self-service by autonomously executing work requests from the unified queue without requiring continuous host processor involvement. This self-service capability ensures reliable control through the adapter's built-in processing logic while maximizing bandwidth utilization by eliminating host processor bottlenecks in the data path.
Data Source
AI summary
A network adapter includes a network interface that communicates packets over a network, a host interface connected locally to a host processor and to a host memory, and processing circuitry, coupled between the network interface and the host interface, and is configured to receive in a common queue, via the host interface, (i) a processing work item specifying a source buffer in the host memory, a data processing operation, and a first address in the host memory, and (ii) an RDMA write work item specifying the first address, and a second address in a remote memory. In response to the processing work item, the processing circuitry reads data from the source buffer, applies the data processing operation, and stores the processed data in the first address. In response to the RDMA write work item the processing circuitry transmits the processed data, over the network, for storage in the second address.