Heterogeneous Packet Transport via Distributed Switch Segmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing packet-based transport systems face challenges in maintaining efficient communication when transporting packets of a first protocol over a second protocol, particularly in avoiding deadlocks and starvation while optimizing for bandwidth and latency.
Innovation Solution
The implementation of distributed switches that re-packetize packets between different protocols, using techniques such as sequence numbering and separate buffers for different packet types to prevent deadlocks and ensure efficient flow control, allowing for the use of different protocols for localized and long-range communication.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If packets of a first protocol are transported over a second protocol using encapsulation, then communication between different protocols is enabled, but deadlocks and starvation can occur in buffered transport systems
Solution Approach 1:
The patent segments packets into different categories (posted requests, completion packets, non-posted requests) and processes them through separate buffers and channels. This segmentation prevents deadlocks by ensuring that critical packets like completion packets can bypass blocked channels and be processed independently, breaking potential circular wait conditions that cause deadlocks in heterogeneous protocol transport.
Solution Approach 2:
The patent introduces an intermediary buffer system that mediates between different protocol packets. The buffer acts as a mediator that can hold and manage packets from the first protocol when being transported over the second protocol, allowing flow control and preventing deadlock conditions by decoupling the transmission and reception processes.
2Productivity
If buffers are used to temporarily hold packets for rate matching, then flow control is achieved, but buffers may become unpractically large for significant rate differences
Solution Approach 1:
The patent divides the buffer system into multiple specialized buffers (posted request buffer, completion packet buffer, non-posted request buffer) rather than using a single large buffer. Each buffer is optimized for specific packet types and can be managed independently, allowing effective flow control with smaller, more manageable buffer sizes that accommodate rate differences without requiring unpractically large storage.
Solution Approach 2:
The patent implements dynamic buffer management where buffers can be selectively filled and emptied based on packet type and system state. The buffer allocation and processing priorities are dynamically adjusted to handle rate matching efficiently, allowing the system to adapt to varying data rates without requiring excessive buffer capacity.
3Loss of energy
If packets are aggregated into larger packets of the second protocol, then network bandwidth efficiency is improved, but latency increases due to additional processing
Solution Approach 1:
The patent segments the packet aggregation process by creating separate processing paths for different packet types. Posted requests can be aggregated and sent efficiently, while completion packets and non-posted requests are handled through separate buffers that allow for lower-latency processing when needed. This segmentation enables the system to optimize for bandwidth efficiency in bulk data transfer while maintaining lower latency for time-sensitive packets.
Solution Approach 2:
The patent applies different quality levels of processing to different packet types. Critical packets like completion packets receive preferential treatment with faster processing and lower aggregation requirements, while less time-sensitive data can undergo more aggressive aggregation for bandwidth efficiency. This local quality approach allows the system to achieve good bandwidth utilization without uniformly increasing latency across all packet types.
4Reliability
If the first protocol is loss-less and requires resending invalid packets, then data integrity is maintained, but communication overhead and latency increase
Solution Approach 1:
The patent implements feedback mechanisms where the receiving entity provides status information about packet reception and validation to the transmitting entity. This feedback allows the system to identify and retransmit only the specific invalid packets rather than resending entire sequences, maintaining data integrity while minimizing retransmission overhead and latency through targeted retransmission based on feedback information.
Data Source
AI summary
For secure transport, when receiving a plurality of packets from a root complex where contents of each packet from the plurality of packets organized in accordance with a first protocol, a sequence number is added to each packet and a packet type is identified. Every packet in the first plurality of packets is encrypted and encapsulated into at least one packet organized in accordance with a second protocol to form a second plurality of packets organized in accordance with the second protocol. All the packets from the second plurality of packets are sent via a plurality of connections so that each connection from the plurality of connections only transports packets from the second plurality of packets that encapsulate packets from the first plurality that have a same packet type.