Network Device Optimizes Distributed Shared Memory Copying
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Distributed Shared Memory (DSM) systems face challenges in managing excessive data copying between nodes, leading to increased network bandwidth usage and performance issues, which are difficult to troubleshoot and require laborious application-specific tuning.
Innovation Solution
A network device identifies excessive data copying in DSM systems and adjusts the data copied, the timing of the copy, and the destination node, using analytics to reduce data copying operations, bandwidth usage, and memory requirements by optimizing data placement across nodes.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If data is copied back and forth between nodes according to access patterns, then data availability is improved, but network bandwidth usage increases
Solution Approach 1:
The system performs preliminary actions by predicting future data access patterns using machine learning models and proactively copying data to appropriate nodes before actual access occurs. This reduces the need for reactive data copying, thereby improving data availability while minimizing network bandwidth consumption during actual access operations.
Solution Approach 2:
The system implements feedback mechanisms by continuously monitoring actual data access patterns and comparing them with predicted patterns. This feedback is used to retrain and refine machine learning models, improving prediction accuracy over time and optimizing data placement decisions to balance data availability with network bandwidth usage.
2Productivity
If page size is reduced to reduce excessive page copying, then data copying operations are reduced, but network message count increases
Solution Approach 1:
The system dynamically changes the parameter of page size based on predicted access patterns and data characteristics. Instead of using fixed page sizes, the system adjusts page granularity adaptively - using larger pages for data with sequential access patterns and smaller pages for data with random access patterns - thereby optimizing the balance between reducing data copying operations and minimizing network message count.
3Productivity
If application-specific tuning is performed to mitigate data copying, then performance is improved, but troubleshooting complexity increases
Solution Approach 1:
The system implements self-service by automatically performing optimization actions based on predicted access patterns without requiring manual application-specific tuning. The machine learning models autonomously analyze access patterns, predict future behavior, and execute data placement optimizations, thereby improving performance while eliminating the need for complex manual troubleshooting and configuration.
Data Source
AI summary
Examples described herein relate to a network interface device. In some examples, the network interface device includes a device interface; input/output circuitry to receive Ethernet compliant packets and output Ethernet compliant packets; circuitry to monitor a particular page for a rate of data copying among nodes within a group of two or more nodes; and circuitry to perform one or more actions based, at least in part, on the rate of data copying among the nodes within the group of two or more nodes to attempt to reduce a number of copy operations of the data among the nodes within the group of two or more nodes, wherein the group of two or more nodes are part of a distributed shared memory (DSM).