Throttle Mechanism for Balancing Host-Synchronous I/O Response Times
Solution Overview
Problem
In data storage systems using Remote Data Facility (RDF) techniques, balancing response times for synchronous I/O requests with different priorities is challenging due to resource starvation caused by intervening non-host synchronous I/O requests, leading to congestion and impact on host response times.
Innovation Solution
Implementing a throttle mechanism that initiates a time window for host-synchronous I/O requests, throttling processing of non-host synchronous I/O requests by limiting resources or generation during this window, with the window length determined by conditions such as network conditions, time of day, or I/O request patterns.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of time
If synchronous I/O requests are processed in priority order, then host response time is improved, but non-host synchronous I/O requests experience resource starvation
Solution Approach 1:
The patent segments synchronous I/O requests into different categories (host-synchronous and non-host-synchronous) and applies different processing strategies to each segment. Host-synchronous requests receive prioritized processing during time windows, while non-host-synchronous requests are throttled during these windows but processed during intervals between windows, resolving the resource starvation issue through segmented handling.
Solution Approach 2:
The patent implements periodic time windows during which host-synchronous I/O requests are prioritized. By alternating between these prioritization windows and intervals where non-host-synchronous requests can be processed, the system ensures both high-priority host responses and fair resource allocation to non-host requests, preventing permanent resource starvation.
2Loss of time
If non-host synchronous I/O requests are throttled during time windows, then host response time is improved, but overall throughput is reduced
Solution Approach 1:
The patent dynamically adjusts the throttling strategy by varying the duration and frequency of time windows based on system conditions. The throttle mechanism is not static but adapts to maintain optimal host response times while minimizing throughput impact, allowing flexible balancing of competing performance goals.
Solution Approach 2:
The patent changes parameters such as time window duration, throttling intensity, and priority levels based on system state. By adjusting these parameters dynamically, the system optimizes the trade-off between host response time and overall throughput, ensuring that throttling does not cause excessive throughput reduction while still achieving improved host response performance.
Data Source
AI summary
A computing environment, such as an data mirroring or replication storage system, may need to process synchronous I/O requests having different priorities in addition to handling I/O requests on the basis of synchronous or asynchronous groupings. The system described herein provides a data storage system that addresses issues involving efficient balancing of response times for servicing synchronous I/O requests having different priorities. Accordingly, the system described herein provides for maintaining an optimal response time for the host-synchronous I/O requests and the optimal throughput of non-host-synchronous I/O requests using a host-synchronous request time window within which processing of non-host-synchronous I/O requests is throttled. The host-synchronous request time window may be selected to enable the optimal response time for the host-synchronous I/O and also to minimize the impact on the overall throughput of the I/O processor of the storage device.