Intel Plumbing the Depths of Cloud Storage: New Opportunities For IO Tracing and Analysis

Intel Labs’ I/O hinting technology (Differentiated Storage Services) is described, along with the new research opportunities it brings, Storage Analytics. The goal of I/O hinting is to bridge the widening semantic gap between computer system and storage systems. By classifying I/O, different classes of data can be handled with different storage system policies. Specifically, when an storage system is first initialized , performance policies are assigned to predefined classes, such as the file system journal. Then, online, a classifier is include with each I/O request, thereby allowing the storage system to enforce the associated policy for each I/O that it receives. Prototypes have been built for various stacks, including file systems, databases, cloud storage stacks, and hypervisors. Once instrumented with I/O classification, stacks can be analyzed in various ways, including bottleneck and dependency analysis, cache utilization, and performance interference among VMs.