Hybrid Cache Coherency Protocol for Multiprocessor Systems
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Multiprocessor computer systems with shared memory architectures face performance drawbacks in cache coherency protocols, including high latency and network congestion due to the need for ordered access rights and acknowledgment messages in directory-based protocols, and bandwidth limitations in broadcast snooping protocols.
Innovation Solution
A hybrid system that dynamically switches between point-to-point and broadcast modes for coherence requests based on network conditions, using a mode control unit and mode table to determine the transmission mode for each coherence request, allowing for efficient cache coherency management without the overhead of directory-based protocols.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If a directory-based cache coherency protocol is used, then cache coherency can be maintained in scalable multiprocessor systems, but latency increases due to the need for ordered access rights and acknowledgment messages
Solution Approach 1:
The system segments the coherency protocol into two independent parts: access right management and data transfer. This allows processors to receive data immediately upon request while access rights are updated separately, eliminating the need to wait for acknowledgment messages before using received data, thus reducing latency while maintaining coherency
Solution Approach 2:
The system performs preliminary data transfer before access right updates are complete. Processors can receive and potentially use data immediately after sending a coherency request, without waiting for the access right changes to be fully propagated through the system. This preliminary action significantly reduces the time processors spend waiting
2Device complexity
If a broadcast snooping cache coherency protocol is used, then simple implementation is achieved, but network bandwidth is limited due to all processors receiving all coherence requests
Solution Approach 1:
The system applies local quality by having the directory selectively forward coherency requests only to processors that have cached copies of the requested data, rather than broadcasting to all processors. This localized approach reduces network traffic while maintaining the simplicity of the broadcast model for processors that don't have the data
3Productivity
If directory-based protocols are used to reduce network traffic, then network congestion decreases, but the overhead of acknowledgment messages and ordered access rights increases latency
Solution Approach 1:
The system segments the coherency protocol into two independent parts: access right management and data transfer. This allows processors to receive data immediately upon request while access rights are updated separately, eliminating the need to wait for acknowledgment messages before using received data, thus reducing latency while maintaining coherency
Solution Approach 2:
The directory acts as an intermediary that manages access rights centrally while allowing direct data transfer between processors. This intermediary approach reduces network traffic by coordinating access rights updates centrally, while the decoupled data transfer eliminates latency-inducing acknowledgment requirements
Data Source
AI summary
A system may include a node and an additional node coupled by an inter-node network. The node includes an active device, an interface to the inter-node network, a memory, and an address network coupling the active device, the interface, and the memory. The active device sends an address packet to initiate a transaction to gain an access right to a coherency unit. In response to receiving the address packet, the memory is configured to send a report corresponding to the address packet to the interface if the transaction cannot be satisfied within the node. The interface is configured to ignore the address packet and to send a coherency message requesting the access right to the additional node via the inter-node network in response to the report.