On Application Responsiveness and Storage Latency in Virtualized Environments

Abstract—Preserving responsiveness is an enabling condition for running interactive applications effectively in virtual machines. For this condition to be met, low latency usually needs to be guaranteed to storage-Input/Output operations. In contrast, in this paper we show that in virtualized environments, there is a missing link exactly in the chain of actions performed to guarantee low storage-I/O latency. After describing this problem theoretically, we show its practical consequences through a large set of experiments with real world-applications. For the experiments, we used two Linux production-quality schedulers, both designed to guarantee a low latency, and a publicly available I/O benchmark suite, after extending it to correctly measure throughput and application responsiveness also in a virtualized environment. Finally, as for the experimental testbed, we ran our experiments on the following three devices connected to an ARM embedded system: an ultra-portable rotational disk, a microSDHC (Secure Digital High Capacity) Card and an eMMC (embedded Multimedia card) device. This is an ideal testbed for highlighting latency issues, as it can execute applications with about the same I/O demand as a general-purpose system, but for power-consumption and mobility issues, the storage devices of choice for such a system are the aforementioned ones. Additionally, the lower the speed of a storage device is, the consequences of I/O-latency are more evident