Abstract: I/O activity of a data storage system is characterized by collecting system traces generated during the I/O activity, grouping records in the system traces according to stores, identifying I/O activity in streams corresponding to the stores, and processing the groups of records to characterize I/O activity for the streams. The stores represent units of storage such as logical drives, single data storage devices and groups of data storage devices in the data storage system. Characterizations of the I/O activity may be used to re-allocate data across the data storage system.

Abstract: The value of a p-quantile of a workload distribution is computed to determine whether a data storage system satisfies a quality-of-service guarantee desired by a client. The quality-of-service guarantee may be a predetermined percentage of a workload being serviced by a device within a predetermined response time. A p-quantile value is set to be equal of the predetermined percentage of the quality-of-service guarantee, and a bound is set to be equal to the predetermined response time. A distribution for the workload arriving at the device during an interval of time equal to the bound is then estimated. It is then determined whether a p-quantile of the workload distribution is less than the bound. In cases where the p-quantile of the workload distribution is computed directly, then the p-quantile of the workload distribution is directly compared to the bound.