Cache Coherency Read Request Phase Reduction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Large multi-processor systems face performance degradation due to increased power consumption and reduced ring bandwidth as the system configuration grows, requiring 5 phases for read requests which increase message traffic and latency.
Innovation Solution
A computer-implemented method reduces the read request phases from 5 to 4 by analyzing replies from nodes in the multi-processor network, allowing either a combined response or data to be sent, thereby optimizing system resources, power consumption, and bandwidth.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If the system uses 5 phases for read requests (request phase, reflected request phase, snoop phase, combined response phase, data transfer phase), then data consistency and correctness are ensured, but message traffic increases and ring bandwidth is reduced
Solution Approach 1:
The patent extracts the data transfer from the traditional 5-phase protocol by allowing the requesting node to obtain data directly from the responding node without requiring the arbiter to combine and retransmit the data. This removes the combined response phase and data transfer phase as separate steps, reducing message traffic while maintaining data consistency through direct peer-to-peer transfer.
Solution Approach 2:
The patent uses the ring network as an intermediary to carry both the request and the response data directly between nodes, eliminating the need for the arbiter to act as a data consolidation intermediary. The ring network itself mediates the data transfer, allowing requests and responses to flow directly through the network without being funneled through the arbiter.
2Reliability
If the arbiter combines snoop replies from all snoopers and sends combined response information on the ring to all snoopers, then cache coherence is maintained, but power consumption increases and ring bandwidth is reduced
Solution Approach 1:
The patent applies local quality by allowing each node to independently process and act on snoop replies relevant to its own cache state without requiring all nodes to receive and process a unified combined response. Each node evaluates the snoop information locally and takes appropriate action, reducing the need for broadcast communication and associated power consumption.
Solution Approach 2:
The patent enables preliminary action by allowing nodes to take cache coherence actions based on snoop replies before a formal combined response is generated and distributed. Nodes can immediately update their cache states or initiate data transfers based on the snoop information they receive, eliminating the waiting period for combined response propagation.
3Reliability
If 5 phases are used for read requests, then system correctness is ensured, but latency increases and system performance degrades
Solution Approach 1:
The patent implements skipping by allowing the system to rush through the traditional 5-phase protocol by combining phases and eliminating unnecessary steps. The requesting node can proceed directly to data transfer upon receiving appropriate snoop replies, skipping the formal combined response phase and reducing overall latency while maintaining correctness through direct validation of cache states.
Solution Approach 2:
The patent merges the combined response phase and data transfer phase into a single operational step where the responding node directly provides data to the requesting node. This consolidation eliminates the sequential overhead of generating a combined response, broadcasting it, and then transferring data, thereby reducing latency while ensuring system correctness through direct data provision.
Data Source
AI summary
A cache coherency technique used in a multi-node symmetric multi-processor system that reduces the number of message phases of a read request from 5 to 4, canceling the combined response phase for read requests in most cases, thereby improving system performance and reducing the overall system power consumption.