Memory Fabric Reliability Zone Endpoint Coordination
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Memory fabrics face challenges in achieving high data reliability with minimal added latency, as existing methods incur significant latency taxes due to serialization, de-serialization, and fabric switching, which impact processor stalling and data availability, especially in scale-out configurations and with limited RAS capabilities in processors supporting DDR memory attach.
Innovation Solution
Implementing a memory fabric with a reliability zone comprising multiple fabric-attached memory endpoints, where a single media controller receives a store command, stores data in non-volatile memory, and sends a single acknowledgement upon successful storage across endpoints, utilizing direct connect or multicast array approaches to maintain low latency and high data reliability.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If data is stored across multiple memory endpoints for high reliability, then data reliability is improved, but latency increases due to serialization, de-serialization, and fabric switching
Solution Approach 1:
The system divides the memory storage function across multiple fabric-attached memory endpoints organized in reliability zones, where each endpoint can independently store data. This segmentation allows parallel storage operations to different endpoints, improving reliability while maintaining low latency through concurrent operations rather than sequential processing.
Solution Approach 2:
The media controller is pre-configured with knowledge of reliability zone mappings and endpoint locations. When a store command arrives, the controller can immediately route data to appropriate endpoints without requiring runtime serialization or de-serialization overhead, as the path and destination are predetermined based on the reliability zone configuration.
2Reliability
If multiple fabric-attached memory endpoints are used for redundancy, then data availability is improved, but device complexity increases
Solution Approach 1:
Each fabric-attached memory endpoint serves multiple functions: it acts as both a primary storage device and a backup device for other endpoints in the reliability zone. This multi-functionality reduces the need for dedicated controller logic for each redundancy pair, simplifying the overall system architecture while maintaining high availability through cross-endpoint redundancy.
Solution Approach 2:
The memory endpoints within a reliability zone autonomously manage their own redundancy relationships without requiring complex external coordination. Each endpoint can independently receive and store data, and the system uses a single acknowledgment mechanism that simplifies the coordination overhead, allowing endpoints to self-manage their redundancy roles.
3Reliability
If traditional I/O storage methods are used with multiple endpoints, then data reliability is improved, but access speed decreases significantly
Solution Approach 1:
The system replaces traditional mechanical I/O storage methods with fabric-attached memory endpoints that use serialized connections similar to DRAM interfaces. This substitution maintains memory-type access latencies (two to three orders of magnitude faster than traditional I/O storage) while achieving reliability through multiple endpoints, effectively replacing slow I/O mechanisms with fast memory-like access patterns.
Solution Approach 2:
The system changes the fundamental access parameters by using memory fabric connections with latency characteristics matching DDR memory attach, rather than traditional I/O interfaces. This parameter change in connection type and access protocol enables fast memory-like speeds while maintaining the reliability benefits of multiple storage endpoints through the fabric switching infrastructure.
Data Source
AI summary
A first endpoint comprises a fabric attach point for attachment to a memory fabric, a first media controller, and a first non-volatile memory media. The memory fabric comprises a reliability zone comprising the first endpoint and at least a second endpoint. The first media controller is configured to receive, from at least one processor coupled to the first endpoint via the at least one fabric attach point, a memory fabric store command to store an object in the reliability zone. The first media controller is further configured to store the object in the first non-volatile memory media, to receive from the second endpoint a message indicating that the same object has been stored by the second endpoint, and to send to the at least one processor a single acknowledgement indicating that the at least one object has been stored in both the first and second endpoints of the reliability zone.