Network Interface Parallelization via Flexible Packet Classification
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing networked computer systems face limitations in scaling network workload due to constraints in packet processing, resource management, and communication overhead, particularly in achieving efficient network throughput, latency, and scalability, which are essential for handling diverse workloads such as bulk data movement, transactional uses, and distributed applications.
Innovation Solution
A network system that parallelizes network interfaces and processing by employing a flexible hierarchical classification scheme, optimistic matching, and internal system driver support, allowing for stateful or stateless associations between workload quanta and processing resources, while addressing mutex contention and thread affinity issues through partitionable and virtualizable architecture.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If network interface functionality is treated as a general purpose I/O interface, then the system maintains generality and modularity, but network throughput and processing efficiency deteriorate
Solution Approach 1:
The network interface functionality is segmented into distinct components: a general purpose I/O interface layer and specialized network processing layers. This segmentation allows the system to maintain generality at the I/O interface level while incorporating specialized network processing capabilities through separate functional modules, thereby resolving the contradiction between generality and network throughput efficiency
Solution Approach 2:
The system implements a multi-functional architecture where the network interface can operate in multiple modes: as a general purpose I/O interface for diverse workloads, and as a specialized network interface optimized for network protocols. This universality allows the same hardware resource to serve both general purposes and network-specific optimized functions, maintaining adaptability while improving network throughput when needed
2Productivity
If low level packet processing is segregated to separate hardware entities, then network processing efficiency improves, but device complexity and system integration difficulty worsen
Solution Approach 1:
The patent merges previously separate hardware entities for packet processing into an integrated network interface unit. The general purpose I/O interface and specialized network processing functions are combined into a single unified device, reducing system integration complexity while maintaining packet processing efficiency through internal optimization of the merged structure
3Productivity
If the number of queues between network interface and system is increased, then network throughput improves, but packet arrival ordering and processing coordination complexity worsen
Solution Approach 1:
The system implements dynamic queue management where the number and configuration of queues are not fixed but can be dynamically adjusted based on workload characteristics. This dynamic approach allows the system to optimize throughput by creating multiple queues when needed while simplifying queue management by reducing to fewer queues when workloads are lighter, thereby adapting to different operational scenarios without permanent complexity
4Productivity
If multiple processors service a network interface, then processing capacity and throughput improve, but coordination overhead and latency worsen
Solution Approach 1:
When multiple processors service the network interface, the system segments the processing workload into distinct functional areas, with each processor handling specific packet processing tasks. This segmentation reduces coordination overhead by minimizing the need for processors to communicate and synchronize, as each processor operates independently on its assigned segment of the workload, thereby maintaining high processing capacity while reducing coordination latency
Data Source
AI summary
A method for scaling a network system which includes providing at least one network interface and providing a flexible association between packets and a plurality of processing entities via the plurality of memory access channels. Each network interface including a plurality of memory access channels.