Disaggregated Memory Appliance Dynamic Allocation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In large datacenter configurations, it is challenging to efficiently provision CPU, memory, and persistent memory resources, leading to stranded resources due to over-provisioning of memory, and increasing difficulties in adding DRAM memory without compromising latency and interconnect bandwidth.
Innovation Solution
A disaggregated memory appliance system comprising leaf memory switches, a low-latency memory switch for connecting external processors to leaf memory modules, and a management processor for configuration and provisioning, enabling dynamic allocation of memory resources while maintaining low latency and high interconnect bandwidth.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Quantity of substance
If memory is over-provisioned in traditional server configurations, then memory availability is improved, but resource utilization deteriorates due to stranded memory
Solution Approach 1:
The patent segments memory resources from individual servers into a disaggregated memory pool that can be dynamically allocated to multiple servers. Memory modules are physically separated from server enclosures and connected via high-speed interconnects, allowing memory to be divided and assigned based on actual demand rather than being permanently bound to specific servers.
Solution Approach 2:
The system implements dynamic memory provisioning where memory allocation is not fixed but can be adjusted in real-time based on workload demands. The disaggregated memory architecture allows memory capacity to be dynamically scaled and reallocated among servers without physical reconfiguration, enabling both high availability and efficient utilization.
2Quantity of substance
If more DRAM modules are added to server systems, then memory capacity is improved, but interconnect bandwidth and latency deteriorate due to bus loading
Solution Approach 1:
The patent extracts memory modules from the traditional server enclosure and places them in separate disaggregated memory appliances. This physical separation removes the memory load from server interconnects, allowing servers to access memory through dedicated high-speed links without competing for bandwidth with other memory operations in the same system.
Solution Approach 2:
The system transitions from a two-dimensional memory architecture (memory within single server enclosures) to a three-dimensional architecture where memory exists in a separate dimension as disaggregated appliances. This spatial reorganization allows memory capacity to scale independently without proportionally increasing interconnect complexity or latency within individual servers.
3Quantity of substance
If memory is added within server enclosures, then memory capacity is improved, but physical space availability deteriorates
Solution Approach 1:
The patent extracts memory modules from constrained server enclosures and houses them in dedicated disaggregated memory appliances with optimized form factors. This extraction frees up valuable server rack space for compute workloads while allowing memory capacity to be expanded in separate, purpose-built infrastructure.
4Adaptability or versatility
If disaggregated memory architecture is implemented, then memory allocation flexibility is improved, but system complexity increases due to management overhead
Solution Approach 1:
The patent implements a universal management interface that handles multiple memory provisioning scenarios through a single standardized system. The disaggregated memory architecture provides multi-functional capabilities including dynamic allocation, capacity expansion, and failover management through a unified control plane, reducing the complexity that would otherwise arise from managing separate memory systems.
Data Source
AI summary
Example embodiments provide a disaggregated memory appliance, comprising: a plurality of leaf memory switches that manage one or more memory channels of one or more of leaf memory modules; a low-latency memory switch that arbitrarily connects one or more external processors to the plurality of leaf memory modules over a host link; and a management processor that responds to requests from one or more external processors for management, maintenance, configuration and provisioning of the leaf memory modules within the memory appliance.