Disaggregated Memory Page Reassignment for Target Performance Ratios
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The construction of large capacity servers with high CPU cores and memory is costly, and existing cloud systems face challenges in efficiently managing memory resources in disaggregated memory structures to meet contractual performance requirements.
Innovation Solution
A method and apparatus for managing disaggregated memory using a memory hypervisor that assigns memory pages between local and remote memories, predicts local memory size based on LRU distance-based histograms, and reassigns pages to ensure target performance ratios are met, enhancing memory efficiency and contractual performance.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Quantity of substance
If large capacity servers with high CPU cores and memory are constructed to provide large capacity cloud system, then memory capacity and computing power are improved, but construction cost increases
Solution Approach 1:
The patent segments memory resources into local memory (attached to computing devices) and remote memory (distributed storage). Instead of requiring each server to have large capacity memory, the system divides memory functionality across multiple components, allowing commodity servers to be used while achieving large capacity through aggregation of distributed memory resources.
Solution Approach 2:
The patent transitions from a single-dimension memory architecture (local memory only) to a two-dimension architecture by adding remote memory dimension. Memory pages can be assigned to either local or remote memory based on access patterns, creating a hierarchical memory space that expands capacity without proportionally increasing cost.
2Ease of manufacture
If disaggregated memory structure is used to support large capacity memory using commodity servers, then construction cost is reduced, but memory access performance deteriorates
Solution Approach 1:
The patent applies local quality by differentiating between frequently accessed memory pages (assigned to local memory for fast access) and infrequently accessed pages (assigned to remote memory). The memory management system analyzes access patterns and dynamically assigns pages to optimize the balance between access speed and resource utilization, ensuring hot data resides in fast local memory while cold data uses cheaper remote storage.
Solution Approach 2:
The patent implements dynamic memory page assignment where the system continuously monitors memory access patterns and reassigns pages between local and remote memory based on changing workload requirements. This dynamic adjustment allows the system to adapt to varying performance demands while maintaining cost efficiency, rather than using a static allocation strategy.
3Productivity
If memory pages are assigned to remote memory to reduce local memory usage, then memory utilization is improved, but access performance deteriorates when target performance ratio is not satisfied
Solution Approach 1:
The patent incorporates feedback mechanisms where the memory management system continuously monitors memory access performance and compares it against target performance ratios. When performance degradation is detected due to remote memory access, the system receives feedback and adjusts page assignment decisions accordingly, migrating pages back to local memory to restore performance while maintaining overall utilization efficiency.
Solution Approach 2:
The patent performs preliminary actions by pre-assigning frequently accessed memory pages to local memory based on predicted access patterns before performance degradation occurs. The system analyzes historical access data and proactively positions likely-to-be-accessed pages in fast local memory, preventing performance issues before they arise rather than reacting after performance deteriorates.
Data Source
AI summary
Disclosed is a method of managing a disaggregated memory. According to the present disclosure, the method includes: assigning at least one memory page to a local memory and a remote memory; checking a request for access to the memory page; checking whether a target performance ratio required in service is satisfied or not when the memory page requested to be accessed is assigned to the remote memory; predicting a size of the local memory on the basis of an LRU distance-based histogram when the target performance ratio is not satisfied; and reassigning the memory page requested to be accessed in consideration of the predicted size of the local memory.