PCI Flow Control via Signal Interception for Higher Bandwidth
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The large round-trip delay for off-chip communications with PCI devices restricts the available bandwidth due to processor elements' tight restrictions on the number of in-flight store with return instructions, leading to inefficient data transmission.
Innovation Solution
Implementing a Device Credit Management Mechanism (DCMM) that intercepts incoming signals to PCI devices, applies flow control, and responds with either acceptance or refusal, acting as a proxy to reduce the round-trip delay and increase the number of in-flight instructions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If store with return instructions are used to send data to PCI devices, then data transmission is achieved, but the large round-trip delay and tight restrictions on in-flight instructions severely restrict bandwidth
Solution Approach 1:
The patent introduces an intermediary mechanism (DCMM - Device Credit Management Module) that sits between the processor element and PCI device. This intermediary intercepts store with return instructions, provides immediate acknowledgment to the processor element, and separately manages data transmission to the PCI device. By decoupling the acknowledgment signal from the actual data transfer, the round-trip delay is eliminated for the processor element, while the intermediary buffers and forwards data to the PCI device at an appropriate rate, thereby resolving the bandwidth restriction caused by the original tight coupling.
2Productivity
If the number of in-flight store with return instructions is restricted, then processor element control is maintained, but bandwidth availability is severely limited
Solution Approach 1:
The patent segments the store with return instruction handling into two independent parts: (1) the instruction issuance and acknowledgment path handled by the DCMM intermediary, and (2) the actual data transmission path to the PCI device. This segmentation allows the processor element to issue instructions without waiting for PCI device response, while the DCMM independently manages the transmission queue and flow control. The segmentation effectively increases bandwidth without requiring complex flow control at the processor element level.
3Productivity
If round-trip delay is reduced through signal interception, then more instructions can be issued over time, but the system requires additional interception and forwarding circuitry
Solution Approach 1:
The DCMM intermediary is designed to perform multiple functions within a single integrated module: it intercepts store with return instructions, generates immediate acknowledgment signals, buffers data, manages flow control credits, and forwards data to PCI devices. By consolidating these multiple functions into a single universal module, the patent reduces the overall system complexity compared to having separate dedicated circuitry for each function, while still achieving the goal of reduced round-trip delay and increased instruction issuance rate.
Data Source
AI summary
A data processing apparatus includes interception circuitry for intercepting an incoming signal corresponding to an instruction from a processor element to a PCI device. Respond circuitry provides a response to the incoming signal back to the processor element and the response is either an acceptance of the incoming signal or a refusal of the incoming signal based on a flow control between the data processing apparatus and the PCI device. Forward circuitry performs a transmission, to the PCI device, of an outgoing signal corresponding to the command after the response has indicated acceptance of the incoming signal.