Towards fairness and efficiency in storage systems
Ajay Gulati, Peter Varman, Arif Merchant, Mustafa Uysal · ACM SIGMETRICS Performance Evaluation Review · 2007
Fairness and overall I/O efficiency are two opposing forces when it comes to sharing I/O among different applications. Although providing QoS guarantees for applications sharing a storage server are desirable under many scenarios, existing work has not been able to make a convincing case for using fairness mechanisms for disk scheduling, mainly due to their impact on overall throughout. In this work, we plan to investigate two major issues: (1) study the trade-off between fairness and efficiency, and develop mechanisms to improve the I/O efficiency of fair schedulers (2) provide performance guarantees to applications in terms of higher-level application metrics (such as transactions/sec), by changing the parameters in a fairness algorithm that affect the allocations at the block level.