I/O Scheduler Synchronization for Disk Latency Reduction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In data management systems, independent I/O schedulers across multiple disks often lead to asynchronous resource deposition, resulting in disk idle times and increased latency due to non-synchronized execution of interactive and non-interactive jobs, which affects the efficiency and throughput of read and write operations.
Innovation Solution
Implementing a synchronization mechanism across I/O schedulers to align resource deposition periods, using a global clock to ensure that non-interactive jobs are executed simultaneously, thereby reducing disk idle times and enhancing the completion of interactive jobs.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If independent I/O schedulers are used across multiple disks, then each disk can autonomously manage its own I/O operations, but resource deposition becomes asynchronous leading to disk idle times and increased latency
Solution Approach 1:
The patent merges the resource deposition timing across multiple independent I/O schedulers by introducing a global clock mechanism. Each I/O scheduler independently deposits resources based on its local clock, but the system synchronizes these deposition events to occur simultaneously across all disks using the global clock reference, thereby eliminating asynchronous idle periods while preserving autonomous management capabilities
Solution Approach 2:
The system implements feedback through the global clock mechanism that monitors and coordinates resource deposition timing across all I/O schedulers. The global clock provides timing feedback to each scheduler, enabling them to align their resource deposition events and coordinate their operational cycles, thus reducing disk idle time while maintaining independent operation
2Productivity
If non-interactive jobs are executed based on local resource availability, then each disk can efficiently utilize its own resources, but execution timing becomes unsynchronized across disks affecting overall job completion
Solution Approach 1:
The patent combines the execution timing of non-interactive jobs across multiple disks by synchronizing resource deposition events. Each disk continues to execute non-interactive jobs based on local resource availability, but the resource deposition events themselves are synchronized to occur simultaneously across all disks using the global clock, ensuring that job execution timing is coordinated without sacrificing local resource utilization efficiency
Solution Approach 2:
The system creates equipotentiality in timing across all I/O schedulers by using the global clock to establish a common reference frame for resource deposition. This ensures that all disks operate at the same temporal potential, depositing resources and executing non-interactive jobs in synchronized time slots, thereby reducing overall job completion time while maintaining efficient local resource usage
3Speed
If interactive jobs are prioritized unconditionally, then interactive performance is maximized, but non-interactive job execution is excessively throttled reducing overall system throughput
Solution Approach 1:
The patent implements periodic action by introducing regular resource deposition events synchronized across all disks using the global clock. Non-interactive jobs are executed during these periodic resource deposition windows, while interactive jobs receive unconditional priority between deposition events. This periodic structure ensures that non-interactive jobs get regular execution opportunities without excessively throttling them, thereby improving overall system throughput while maintaining interactive performance
Solution Approach 2:
The system applies dynamics by making resource availability time-dependent through the synchronized deposition mechanism. Resource availability for non-interactive jobs dynamically changes at regular intervals defined by the global clock, creating windows of opportunity for non-interactive execution. This dynamic resource allocation allows the system to flexibly balance between interactive and non-interactive workloads, improving overall throughput without compromising interactive job speed
Data Source
AI summary
Methods, systems, and devices for data management are described. A data management system (DMS) may include multiple disks. A distributed file system may be stored across the multiple disks. The DMS may receive a request to perform a first job for the distributed file system. The DMS may send the first job for execution, during a time period, by a set of disks. The first job may correspond to input/output (I/O) operations associated with a first priority that is higher than second priorities associated with second jobs for execution by the set of disks. Respective timings for executing the second jobs may be based on respective availabilities of resources at the set of disks. The DMS may synchronize, during the time period, the respective availabilities of resources across the set of disks based on the set of disks executing the first job associated with the first priority.