Polling-based I/O optimization | Optimization of Polling-based I/O | Android Polling based I/O optimization

When increasing number of CPU cores and decreasing I/O latency of a block device have led to a rediscovery of the value of polling-based I/O . State-of-the-art smartphones contain quad-core CPUs and NAND-based storage latency is an order of magnitude smaller than that of legacy hard disk drive(HDD). In this environment, interrupt-driven I/O may hinder the performance of a system due

to context switches. When many small I/Os are generated from the block I/O layer, the I/O daemon for the eMMC, mmcqd, is subject to significant context switch overhead. Our results below show that this can indeed be the case and also show that polling-based I/O can provide a superior I/O performance to interrupt driven one without sacrificing the overall system performance. We modify the I/O subsystem for the Android platform so that the mmcqd uses polling to access the stor-

age device. There are two issues in polling-based I/O , CPU monopolization and power consumption. We perform an experiment if the polling based I/O interferes with the ongoing application, articulately CPU intensive one. We ran a HD-quality (1920x1080 at 30 fps) video recording application concurrently with our benchmark process.