Hypervisor I/O Access Suppression for Fault-Tolerant SoC Performance
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Fault-tolerant (FT) servers without hypervisors face performance degradation and configuration limitations due to the need for physical separation of CPU subsystems and input/output devices, making them unsuitable for System on Chip (SoC) configurations, while hypervisor-based FT servers incur overhead from virtualization, degrading input/output access performance.
Innovation Solution
An information processing apparatus with a virtual machine, hypervisor, monitoring unit, notification unit, and suppression unit that detects data output attempts during synchronization processes, allowing direct operation of physical devices and avoiding unnecessary synchronization, thus enhancing performance and supporting SoC configurations.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If a hypervisor is used to manage FT server, then system reliability is improved through virtualization, but input/output access performance deteriorates due to virtualization overhead
Solution Approach 1:
The patent segments the FT server management into two parts: the hypervisor handles high-level virtualization and reliability management, while a dedicated driver in the virtual machine handles specific I/O device operations. This segmentation allows the driver to bypass hypervisor overhead for I/O operations, improving performance while maintaining the hypervisor's reliability functions.
Solution Approach 2:
The patent introduces a driver as an intermediary component between the virtual machine and the I/O device. This driver acts as a mediator that enables direct I/O access without requiring full hypervisor intervention, thus reducing virtualization overhead while maintaining system reliability through the hypervisor's ongoing management.
2Reliability
If physical separation of CPU subsystems and input/output devices is implemented, then fault tolerance is improved, but device complexity increases due to additional components
Solution Approach 1:
The patent makes the I/O device accessible to multiple CPU subsystems through a unified interface. The device can be shared between primary and secondary CPU subsystems, eliminating the need for separate dedicated devices for each subsystem while maintaining fault tolerance through the hypervisor's coordination.
Solution Approach 2:
The patent uses virtualization to create a virtual representation of the I/O device that can be accessed by multiple CPU subsystems. This virtual copy allows logical separation and sharing without requiring physical duplication of devices, reducing complexity while maintaining fault tolerance.
3Reliability
If I/O device is virtualized by hypervisor, then system reliability is improved through centralized control, but input/output access performance deteriorates due to overhead
Solution Approach 1:
The patent segments I/O control into two layers: the hypervisor maintains virtualization and reliability management, while a dedicated driver in the virtual machine handles actual I/O operations. This segmentation allows direct hardware access for performance-critical operations while preserving the hypervisor's reliability functions.
Solution Approach 2:
The patent enables the virtual machine's driver to directly manage I/O device operations without continuous hypervisor intervention. The driver performs self-service I/O operations, reducing the overhead of hypervisor-mediated access while the hypervisor maintains overall system reliability through periodic checks and coordination.
Data Source
AI summary
An information processing apparatus includes: a VM including a driver that is configured to perform data input/output using an input/output device; a hypervisor configured to manage the VM and perform a synchronization process between the information processing apparatus and another information processing apparatus which form a pair; a monitoring unit configured to monitor an access made between a memory and the input/output device; a notification unit configured to notify the hypervisor of detection of a reading access in a case where the reading access relating to data output from the memory to the input/output device is detected by the monitoring unit; and a suppression unit configured to suppress the data output in a case where the hypervisor is in the middle of execution of the synchronization process when the hypervisor receives the notification from the notification unit. Accordingly, the input/output access performance of the VM is improved.