Look-Aside RAID Parity for SSD Controller Bottlenecks
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
RAID systems using solid state drives face performance bottlenecks due to the overhead of write-back mode operations, which overwhelm the RAID controller and degrade the reduced latency offered by PCIe interfaces.
Innovation Solution
Implementing look-aside parity by mirroring information writes to both a storage array controller and storage devices, with the switching device generating parity information and distributing RAID operations between the RAID controller and the switching device, such as a PCIe switch, to reduce congestion and latency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If write-back mode operations are used in RAID systems with solid state drives, then data redundancy and reliability are improved, but the RAID controller becomes overwhelmed and system performance degrades
Solution Approach 1:
The patent segments the RAID functionality by separating parity generation and writing operations from the RAID controller. The host system directly writes data to storage devices while a separate parity generation component handles parity calculations, dividing the workload to prevent controller overload and maintain both reliability and performance.
Solution Approach 2:
The patent introduces an intermediary parity generation component that acts as a mediator between the host system and storage devices. This intermediary handles the computationally intensive parity operations independently, allowing the RAID controller to focus on data management while maintaining data redundancy without performance degradation.
2Productivity
If write-through mode is used with solid state drives, then RAID controller overhead is reduced, but the reduced latency advantage of PCIe interfaces and solid state drives is degraded
Solution Approach 1:
The patent implements preliminary action by having the host system perform data writing operations directly to storage devices before parity generation occurs. This allows data to be committed to storage immediately, utilizing the low latency of PCIe interfaces and solid state drives, while parity generation proceeds separately without blocking the write path.
3Reliability
If parity operations are performed by the RAID controller, then data redundancy is maintained, but congestion and bottlenecks occur at the RAID controller
Solution Approach 1:
The patent extracts the parity generation and writing operations from the RAID controller and relocates them to separate components. The RAID controller retains responsibility for maintaining data redundancy through coordinated parity management, while the computationally intensive parity operations are performed independently, eliminating congestion at the controller.
4Reliability
If mirroring is used to maintain exact copies of storage devices, then data reliability is improved, but storage device space efficiency decreases
Solution Approach 1:
The patent changes the redundancy parameter from full mirroring (100% duplicate data) to parity-based redundancy (distributed checksum data). This parameter change allows the system to maintain data reliability while significantly improving storage space efficiency, as parity information requires less space than complete data copies and can be distributed across multiple devices.
Data Source
AI summary
Redundant storage of information is provided by distributing storage functions between a RAID controller and switching device. The switching device multi-casts writes to storage devices and to the RAID controller. The RAID controller generates parity for the information and writes the parity to the storage devices in space reserved for parity by the switching device. Information is read from the storage devices through the switching device without action by the RAID controller.