Flash Drive I/O Scheduling for Read-Write Separation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing data storage systems employing solid state storage devices face performance issues due to differing read/write performance dynamics compared to spinning plate track/sector based disk drives, leading to suboptimal I/O performance in RAID configurations like RAID 1, RAID 5, and RAID 6.
Innovation Solution
Implementing a scheduling policy that designates specific solid state storage devices as either current write devices or current read devices, swapping roles based on triggering events to maximize performance by keeping each device close to either 100% reads or 100% writes, thereby optimizing I/O operations.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If traditional RAID scheduling algorithms are used for solid state storage devices, then data access performance is optimized for spinning plate track/sector based disk drives, but read/write performance dynamics of solid state storage devices are not properly addressed, resulting in suboptimal I/O performance
Solution Approach 1:
The patent changes the scheduling parameters from those optimized for spinning disk drives to those optimized for solid state storage devices. Specifically, it transitions from seek-time and track/sector-based scheduling to I/O operation type-based scheduling, where operations are classified as read-intensive or write-intensive and directed to appropriate devices based on their current performance state and operation type.
Solution Approach 2:
The patent implements dynamic scheduling that adapts to the real-time performance characteristics of solid state storage devices. The system continuously monitors device performance and dynamically adjusts I/O operation assignment, switching between read-intensive and write-intensive modes based on current device state, thereby optimizing I/O performance for varying workloads.
2Productivity
If solid state storage devices perform both reads and writes simultaneously, then device utilization is maximized, but read/write interference occurs, reducing overall device performance
Solution Approach 1:
The patent segments I/O operations into distinct categories: read-intensive operations and write-intensive operations. By classifying and separating these operation types, the system can direct them to appropriate storage devices based on current device performance state, thereby avoiding read/write interference while maintaining high device utilization through specialized operation assignment.
Solution Approach 2:
The patent applies local quality by assigning different operation types to different devices based on their current performance characteristics. Rather than treating all devices uniformly, the system identifies devices currently optimized for reads versus writes and selectively routes I/O operations accordingly, ensuring each device operates in its optimal performance mode.
3Productivity
If storage devices are frequently swapped between read and write roles, then performance optimization is achieved, but system complexity and swap overhead increase
Solution Approach 1:
The patent implements periodic role swapping between storage devices, where devices alternate between read-intensive and write-intensive roles at predetermined intervals or based on performance thresholds. This periodic action simplifies the scheduling logic compared to continuous dynamic adjustment, as devices follow a predictable rotation pattern that reduces computational overhead while maintaining performance optimization.
Data Source
AI summary
In a data storage system, a controller schedules I/Os to storage devices so that each one substantially performs only reads or only writes, thereby increasing performance. At least one storage device is designated as a current write device. The remainder of the devices are designated current read devices. Host write data is stored in a buffer memory. Storage device reads occur only from the current read devices. Storage device writes occur only to the current write device(s). In response to a triggering event, the designations are updated so that a different storage device is designated the current write device, and the remainder of the storage devices are designated the current read devices. A triggering event may include but not be limited to a time period, number of writes, cache size, device wear, environmental conditions, application requirements, or combination.