Initiator-Side Network Offload for NVMe Scale-Out Storage
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing data center storage systems face inefficiencies in managing block-level storage, particularly with ephemeral storage tied to compute instances, leading to data loss and infrastructure challenges during VM or container migration, and underutilization or over-constraint of local storage capacity.
Innovation Solution
Implementing a remote disaggregated scale-out storage system where the initiator circuitry is offloaded to a network interface device, utilizing programmable circuitry to map NVMe commands to network-accessible storage locations, enabling efficient management of IO operations through RDMA and reliable transport protocols, and using programmable registers to manage exceptions and connections.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Speed
If storage volumes are directly attached to a host that executes a VM or container, then access speed and performance are improved, but storage capacity is underutilized or over-constrained and infrastructure challenges arise during VM or container migration
Solution Approach 1:
The storage system is segmented into direct-attached storage components (for performance) and remote disaggregated storage components (for flexibility and migration). The NVMe-oF protocol enables the host to access remote storage as if it were locally attached, combining the benefits of both approaches by segmenting the storage architecture into different layers with different access characteristics.
2Adaptability or versatility
If remote disaggregated storage architectures are used with storage servers accessed over a network, then storage capacity and flexibility are improved, but software overhead increases due to mapping requirements
Solution Approach 1:
The patent replaces software-based mapping mechanisms with hardware-based NVMe-oF protocol processing. The network interface and storage system use standardized NVMe commands and protocols to directly map and access storage resources, eliminating the need for complex software mapping layers while maintaining flexibility and scalability.
3Reliability
If initiator circuitry is implemented in the host, then storage access control and management are improved, but processing overhead on the host increases
Solution Approach 1:
The initiator circuitry and its processing functions are extracted from the host system and integrated into the storage system itself. The storage devices include embedded controllers that handle NVMe command processing, connection management, and data access control, freeing the host from these overhead tasks while maintaining reliable access control through the standardized NVMe protocol.
Data Source
AI summary
Examples described herein relate to a network interface device that includes circuitry to perform operations, offloaded from a host, to identify at least one locator of at least one target storage associated with a storage access command based on operations selected from among multiple available operations, wherein the available operations comprise two or more: entry lookup by the network interface device, hash-based calculation on the network interface device, or control plane processing on the network interface device.