Memory Channel Network Architecture for Near-Memory Processing
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current server architectures face limitations in compute throughput and memory bandwidth, particularly when dealing with emerging data-intensive applications like big-data analytics, due to constraints in DRAM capacity and DDR bandwidth.
Innovation Solution
The Memory Channel Network (MCN) architecture integrates near-memory processing with distributed computing frameworks, providing a standard, application-transparent communication interface between hosts and near-memory processors, and among servers, by leveraging high-bandwidth and low-latency DDR interfaces.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If near-memory processing architectures are implemented to increase compute throughput and memory bandwidth, then processing performance is improved, but application readiness deteriorates due to significant changes required in target applications
Solution Approach 1:
The patent introduces a communication interface that acts as an intermediary layer between distributed computing frameworks and near-memory processors. This interface translates standard communication protocols into near-memory processing operations, allowing applications to interact with near-memory processors without requiring significant modifications to their code or architecture.
Solution Approach 2:
The communication interface is designed to be universal, supporting multiple distributed computing frameworks (such as MPI, Spark, and Hadoop) through a single standardized interface. This multi-functionality allows the near-memory processing architecture to work with various application types and frameworks without requiring framework-specific implementations.
2Quantity of substance
If memory capacity is increased to handle data-intensive applications, then storage capability is improved, but compute throughput remains limited by DDR bandwidth
Solution Approach 1:
The patent moves processing operations from the traditional CPU-centric dimension to a near-memory dimension, where processing units are physically located adjacent to memory devices. This spatial reorganization creates a new operational dimension that bypasses the DDR bandwidth bottleneck by enabling direct access to memory capacity without requiring data to traverse the traditional memory bus.
3Ease of manufacture
If servers are designed to be cost-effective with increased memory capacity, then economic efficiency is improved, but compute throughput and memory bandwidth must be increased commensurately
Solution Approach 1:
The patent merges memory devices with processing units into integrated near-memory processing modules. This consolidation allows servers to increase memory capacity without proportionally increasing the number of separate memory controllers and DDR interfaces, thereby maintaining cost-effectiveness while achieving higher aggregate memory bandwidth through parallel access to multiple integrated memory devices.
Data Source
AI summary
A computing device includes a host processor to execute a host driver to create a host-side interface, the host-side interface emulating a first Ethernet interface, assign the host-side interface a first medium access control (MAC) address and a first Internet Protocol (IP) address. Memory components are disposed on a substrate. A memory channel network (MCN) processor is disposed on the substrate and coupled between the memory components and the host processor. The MCN processor is to execute an MCN driver to create a MCN-side interface, the MCN-side interface emulating a second Ethernet interface. The MCN processor is to assign the MCN-side interface a second MAC address and a second IP address, which identify the MCN processor as a MCN network node to the host processor.