Reconfigurable Dataflow Unit P2P Protocol
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing communication systems between reconfigurable dataflow units (RDUs) face challenges in efficiently managing peer-to-peer transactions across intra-die and inter-die networks, leading to potential bottlenecks and reduced dataflow bandwidth.
Innovation Solution
The implementation of a peer-to-peer (P2P) communication protocol that utilizes transactions encapsulated in Transaction Layer Packets (TLPs) of PCIe, allowing for direct communication between RDUs without the need for a host intermediary, and supporting various primitive and complex transactions such as remote read/write, stream write, and barrier operations.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If traditional communication protocols are used between RDUs, then communication can be established, but dataflow bandwidth is reduced and bottlenecks occur
Solution Approach 1:
The patent introduces a peer-to-peer communication protocol that acts as an intermediary layer between RDUs, enabling direct communication without host intervention. This mediator protocol operates at the transaction layer, translating between different RDU communication interfaces and enabling high-bandwidth dataflow while maintaining protocol compatibility across diverse RDU architectures.
Solution Approach 2:
The communication system is segmented into distinct layers: the peer-to-peer protocol layer handles high-bandwidth data transfer, while lower layers manage physical interfacing. This segmentation allows the dataflow path to be separated from control paths, enabling sustained high bandwidth without protocol overhead bottlenecks.
2Productivity
If host intermediary is used for RDU communication, then protocol compatibility is maintained, but communication efficiency and bandwidth are reduced
Solution Approach 1:
The patent extracts the communication coordination function from the host processor and embeds it directly into the RDU interconnect fabric. The peer-to-peer protocol extracts essential coordination mechanisms (arbitration, flow control, error handling) from complex host-based protocols, providing RDU-specific optimization while maintaining sufficient compatibility for system operation.
Solution Approach 2:
RDUs are equipped with built-in peer-to-peer communication capabilities that enable them to service their own communication needs without host intervention. Each RDU includes protocol handling logic and communication interfaces that autonomously manage data exchange, freeing the host from communication management overhead and improving overall system efficiency.
3Productivity
If direct peer-to-peer communication is implemented between RDUs, then dataflow bandwidth is improved, but protocol compatibility challenges arise
Solution Approach 1:
The peer-to-peer communication protocol is designed with universal interface capabilities that can adapt to multiple RDU architectures and communication standards. The protocol includes translation and adaptation layers that maintain compatibility with diverse RDU types while providing unified high-bandwidth access, enabling one protocol to serve multiple function and interface requirements.
Solution Approach 2:
The protocol employs configurable parameters that can be adjusted based on the specific RDU pair communicating. These parameters include data width, timing characteristics, and protocol version selection, allowing the same peer-to-peer infrastructure to adapt to different RDU implementations while maintaining high bandwidth performance across varied configurations.
Data Source
AI summary
A reconfigurable processing unit is disclosed, comprising a first internal network and a second internal network with different protocols, an interface to an external network with a different protocol, a first configurable unit connected to the first internal network, a second configurable unit connected to both the first internal network and the second internal network, and a third configurable unit connected to both the second internal network and the interface to the external network. The third configurable unit is configured to receive a payload from the external network and send the transaction type identifier and the source application ID to the second configurable unit over the second internal network. The second configurable unit sends information to the first configurable unit based on the transaction type identifier and the source application ID matching the local application ID retrieved from the register.