Pallavi Joseph Kalpana and Furquan Shaikh
Disk subsystems are increasingly being accessed by multiple workloads with diverse characteristics and requirements. Current disk I/O scheduler design is based upon certain assumptions made about the latency characteristics of the underlying disk technology like electromechanical disks etc. These assumptions do not hold in virtualized environment leading to sub optimal performance using the same scheduling algorithms. This is because the virtual disk latency characteristics differ markedly from the actual ones.
Our project proposes a novel Two-Tier Disk I/O Scheduling technique to be incorporated in the current disk I/O scheduler for DAS design at both the Virtual Machine (VM) level and Virtual Machine Monitor(VMM) level. The Tier-1 design proposes to switch between the existing disks I/O scheduling algorithms at runtime based on the varying workload thus ensuring optimal throughput at VM-level. It also suggests adapting the existing algorithms to consider various characteristics such as disk latency. In Tier-2, the VMM-level disk I/O scheduler works in resonance with the I/ O scheduling performed at Tier-1 of various VMs and accordingly reorders the requests to be served. This Tier-2 design takes the optimization to a further level by maintaining inter-domain fairness and performance isolation. This technique is expected to support the hypothesis of significant performance improvements in Virtualized Environment.