3DS DRAM Memory Mapping for Thermal-Aware VM Allocation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The issue of thermal dissipation and potential thermal runaway in 3D stacked DRAM modules, particularly exacerbated by processor-in-memory modules, leads to degradation in memory performance and bandwidth due to inadequate consideration of individual DRAM die thermal characteristics in current system software mapping approaches.
Innovation Solution
Implementing a system that maps memory allocations to DRAM dies based on their thermal dissipation characteristics, using a BIOS configuration knob to assign thermal scores and allocate VMs and applications to memory ranks with appropriate thermal dissipation, combined with hypervisor/OS-controlled memory chunk interleaving and live VM migration to manage temperature distribution.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If memory allocations are mapped to DRAM dies without considering thermal dissipation characteristics, then memory bandwidth and performance are maximized, but thermal runaway occurs and temperature distribution becomes uneven
Solution Approach 1:
The patent assigns different thermal scores to different memory ranks based on their thermal dissipation characteristics. The system then maps memory allocations to specific ranks based on their thermal properties, ensuring that workloads are distributed according to local thermal quality rather than treating all ranks uniformly. This resolves the contradiction by optimizing memory bandwidth while preventing thermal runaway through location-aware allocation.
Solution Approach 2:
The system dynamically monitors temperature and adjusts memory allocation mappings in real-time. When thermal conditions change, the system can migrate workloads between memory ranks to maintain optimal temperature distribution. This dynamic adaptation allows the system to maintain high memory bandwidth while preventing thermal runaway by responding to changing thermal conditions.
2Productivity
If memory access rate is increased to improve performance, then productivity increases, but thermal dissipation becomes inadequate and thermal runaway occurs
Solution Approach 1:
The patent changes the parameter of memory allocation by introducing thermal scores and mapping workloads to memory ranks based on their thermal dissipation characteristics. Instead of uniformly increasing memory access rates across all ranks, the system adjusts the parameter of workload distribution to match thermal capabilities, allowing high performance while maintaining adequate thermal dissipation.
Solution Approach 2:
The system implements feedback by monitoring temperature conditions and adjusting memory allocation accordingly. When thermal dissipation becomes inadequate, the system receives feedback about temperature levels and modifies workload distribution to prevent thermal runaway. This feedback mechanism enables the system to maintain high productivity while ensuring thermal dissipation remains adequate.
3Reliability
If refresh rate is increased to preserve memory contents at high temperatures, then memory reliability is maintained, but memory performance and bandwidth degrade
Solution Approach 1:
The system performs preliminary action by proactively mapping memory allocations to cool memory ranks before thermal runaway occurs. Instead of reacting to high temperatures by increasing refresh rates, the system prevents the need for high refresh rates by distributing workloads to ranks with adequate thermal dissipation capabilities. This preliminary thermal management maintains both reliability and performance.
Solution Approach 2:
The patent converts the potential harm of thermal accumulation into a benefit by using thermal scores to guide workload distribution. Rather than allowing high temperatures to degrade performance and then compensating with increased refresh rates, the system uses thermal characteristics to optimally place workloads, transforming the thermal challenge into a performance optimization opportunity.
Data Source
AI summary
Methods and apparatus for mapping memory allocation to DRAM dies of a stacked memory modules are described herein. Memory address ranges in a module employing 3DS (three dimensional stacked) DRAMs (Dynamic Random Access Memories) comprising stacked DRAM dies are mapped to DRAM dies in the module based on a layer of the DRAM dies, where dies in different layers have different thermal dissipation characteristic. Chunks of the memory address range are allocated to software entities such as virtual machines (VMs) and/or applications based on a memory access rate of the VMs/applications and the thermal dissipation characteristics of the DRAM die layers, wherein VMs/applications with higher memory access rate are allocated memory on DRAM dies with higher thermal dissipation. In one aspect, memory ranks are associated with respective die layers. In response to detection of change in access rates, memory may be migrated between ranks. Interleaving at multiple levels is also supported.