Dynamic Memory Configuration for Edge Power Efficiency
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current memory subsystems in edge data centers are inflexible and power-inefficient, as they require fixed configurations and power consumption for all memory tiers, which is wasteful in power-constrained environments like edge compute resources powered by ambient sources.
Innovation Solution
Implementing a dynamic memory configuration system that decomposes memory into fine-grained modules that can be turned on or off based on power availability, using hybrid DIMMs with multiple memory technologies and leveraging accelerators for data movement between tiers, and employing RDMA for remote memory access to optimize power consumption while meeting latency requirements.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Stability of the object's composition
If a pre-determined memory configuration is used, then system stability is improved, but power efficiency deteriorates
Solution Approach 1:
The patent implements dynamic memory configuration that allows the system to change memory topology, address maps, and DIMM interleaving patterns at runtime based on power availability and application requirements, transforming the static memory subsystem into an adaptive one that can optimize power consumption while maintaining operational stability
Solution Approach 2:
The system changes key memory parameters including address map configurations, DIMM interleaving patterns, and memory tier activation states based on ambient power conditions and application needs, allowing the same hardware to operate in different configuration modes to balance stability and power efficiency
2Speed
If all memory tiers are always active, then access speed is improved, but power consumption increases
Solution Approach 1:
The memory subsystem is segmented into multiple independent tiers (DRAM, DCPMM, remote memory) that can be selectively activated or deactivated. Each tier can be independently controlled, allowing the system to activate only the necessary memory layers based on application requirements and power availability, thus reducing overall power consumption while maintaining access speed for active tiers
Solution Approach 2:
The system implements periodic monitoring of ambient power conditions and dynamically adjusts memory tier activation states accordingly. When power is abundant, higher-performance memory tiers are activated; when power is constrained, the system transitions to lower-power configurations, creating a rhythmic adaptation pattern that balances performance and energy usage
3Reliability
If memory configuration is fixed at boot time, then system reliability is improved, but adaptability to power conditions deteriorates
Solution Approach 1:
The patent transforms the static memory configuration into a dynamic system that can adapt to changing power conditions while maintaining reliability through controlled transitions. The system monitors power conditions and reliably switches between predefined configuration states, ensuring stable operation regardless of power availability
4Use of energy by moving object
If fine-grained memory modules are used, then power efficiency is improved, but device complexity increases
Solution Approach 1:
The memory subsystem is divided into fine-grained, independently controllable modules and tiers. Each module can be individually activated or deactivated based on power conditions, allowing precise power control. While this increases granularity, the system manages complexity through standardized interfaces and automated configuration management
Data Source
AI summary
Methods and apparatus for platform ambient data management schemes for tiered architectures. A platform including one or more CPUs coupled to multiple tiers of memory comprising various types of DIMMs (e.g., DRAM, hybrid, DCPMM) is powered by a battery subsystem receiving input energy harvested from one or more green energy sources. Energy threshold conditions are detected, and associated memory reconfiguration is performed. The memory reconfiguration may include but is not limited to copying data between DIMMs (or memory ranks on the DIMMS in the same tier, copying data between a first type of memory to a second type of memory on a hybrid DIMM, and flushing dirty lines in a DIMM in a first memory tier being used as a cache for a second memory tier. Following data copy and flushing operations, the DIMMs and/or their memory devices are powered down and/or deactivated. In one aspect, machine learning models trained on historical data are employed to project harvested energy levels that are used in detecting energy threshold conditions.