Interconnection Device Multicast Request Merging
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Multi-processor systems face significant bandwidth bottlenecks in memory access and network link bandwidth consumption due to multiple processors accessing the same data, leading to inefficient data retrieval and transmission processes.
Innovation Solution
An interconnection device with a request processing module that merges data access requests from multiple processors within a multicast group, using a merge bit, multicast group identifier, and transaction identifier to optimize data access and transmission by sending merged requests through the network, reducing the need for repeated data access and transmission.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If multiple processors access the same data from memory individually, then each processor can retrieve data independently, but memory access bandwidth and network link bandwidth consumption increase significantly
Solution Approach 1:
The patent merges multiple identical data access requests from different processors into a single multicast request. The interconnection device identifies that multiple processors (e.g., processors 10i, 10j, 10m, 10n) are requesting the same data from memory 30k, combines these requests into one multicast request, and transmits it through the interconnection network. This reduces memory access bandwidth consumption from 4x to 1x and significantly reduces network link bandwidth consumption while maintaining independent data access capability for each processor.
2Loss of energy
If data access requests are merged into multicast requests, then bandwidth consumption is reduced, but the complexity of request processing and routing increases
Solution Approach 1:
The patent introduces interconnection devices (201, 202, 203) as intermediaries between processors and memory. These intermediary devices perform the complex task of identifying identical requests, merging them into multicast requests, and routing them appropriately. The processors themselves remain simple, issuing standard read requests without needing to implement complex merging logic. This distributes the processing complexity to the interconnection network infrastructure rather than to each processor.
Solution Approach 2:
The interconnection devices perform preliminary analysis of incoming requests to identify which requests are identical and can be merged. By detecting duplicate requests early in the request flow (before they reach the memory), the system can consolidate them into single multicast requests, reducing subsequent processing complexity and bandwidth usage.
3Reliability
If individual read requests are sent from each processor to memory, then data access is straightforward and reliable, but the same data is read multiple times consuming excessive bandwidth
Solution Approach 1:
Instead of having each processor read the same data independently from memory, the patent creates a single copy of the data in the interconnection network's multicast capability. The interconnection device receives one read request from memory, copies the data to multiple processors through the interconnection network's multicast function, and eliminates the need for multiple identical read operations. This maintains data reliability while reducing memory bandwidth consumption from 4x to 1x.
Data Source
AI summary
An information processing method, an interconnection device, and a computer-readable storage medium are provided. The interconnection device includes a request processing module configured for: receiving a data access request from at least one processor, wherein the data access request comprises a merge bit, a multicast group identifier (MGID), and a multicast transaction identifier (MTID); determining whether the data access request is a multicast request; determining whether the interconnection device receives other multicast requests if it is determined that the data access request is a multicast request based on the MGID, the MTID, and a static routing policy of a multicast group; and obtaining the other multicast requests if it is determined that the interconnection device receives the other multicast requests, merging the multicast request with the other multicast requests into a merged request, and forwarding the merged request to a next-hop device of the interconnection device.