Topology Data Structure for Selective Cache Coherence Broadcast
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional cache coherent multiprocessor systems face inefficiencies due to the need for global broadcasts of memory access requests and coherence messages, leading to increased latency and bandwidth requirements as system size scales, as all nodes typically receive memory access requests and coherence messages, even if not necessary.
Innovation Solution
Implementing a topology data structure that allows for intelligent selection of coherence participants for memory access requests and coherence messages, enabling a master to issue requests with a remote scope of broadcast that includes only specific nodes, reducing unnecessary broadcasts and improving efficiency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If global broadcast of memory access requests is used to maintain cache coherence, then all coherence participants receive the requests ensuring coherence, but traffic volume and latency increase as system size scales
Solution Approach 1:
The patent applies local quality by differentiating the broadcast scope based on the target address characteristics. Instead of uniform global broadcasting, the system determines whether to broadcast locally within a node or remotely to specific nodes based on the address translation result. This selective approach reduces unnecessary broadcasts to nodes that don't need the coherence message, thereby reducing latency while maintaining coherence where required.
Solution Approach 2:
The patent segments the broadcast domain into local and remote portions based on the memory topology. By using the topology data structure to identify which nodes actually need to receive coherence messages, the system divides the broadcast traffic into necessary remote broadcasts and excluded nodes, reducing overall traffic volume and latency while preserving cache coherence for the segmented relevant participants.
2Reliability
If global broadcast of memory access requests is used, then cache coherence is maintained across all nodes, but bandwidth consumption increases with system size
Solution Approach 1:
The system applies local quality by making the broadcast scope address-dependent. The topology data structure enables the system to identify which specific remote nodes need to receive coherence messages based on the target address, rather than broadcasting to all nodes uniformly. This reduces bandwidth consumption by excluding nodes that don't need the message while maintaining coherence for the necessary participants.
Solution Approach 2:
The patent segments the system into coherence domains using the topology data structure. By determining which nodes are actually participants in the coherence protocol for a given address, the system divides the broadcast traffic into necessary remote nodes and excluded nodes, thereby reducing overall bandwidth consumption while maintaining coherence where required.
3Loss of time
If restricted scope broadcast is used for memory access requests, then traffic volume and latency are reduced, but cache coherence may be compromised if necessary nodes are excluded
Solution Approach 1:
The topology data structure serves as an intermediary that mediates between the restricted broadcast scope and cache coherence requirements. It translates the target address into specific node identifiers, enabling the system to precisely include only the necessary remote nodes in the broadcast while excluding others. This intermediary mechanism ensures coherence is maintained for the correct participants without the overhead of global broadcasting.
Data Source
AI summary
A coherent data processing system includes a system fabric communicatively coupling a plurality of nodes arranged in a plurality of groups. A plurality of coherence agents are distributed among the nodes and are assigned responsibility for certain addresses. A topology data structure indicates by group and node differing physical locations within the data processing system of the plurality of coherence agents. A master accesses the topology data structure utilizing a request address to obtain a particular group and node of a particular coherence agent uniquely assigned the request address. The master initially issues, on the system fabric, a memory access request specifying the request address and utilizing a remote scope of broadcast that includes the particular node and excludes at least one other node in the particular group, where the particular node is a different one of the plurality of nodes than a home node containing the master.