PERSES: Data Layout for Low Impact Failures

Abstract

Growth in disk capacity continues to outpace advances in read speed and device reliability. This has led to storage systems spending increasing amounts of time in a degraded state while failed disks reconstruct. Users and applications that do not use the data on the failed or degraded drives are negligibly impacted by the failure, increasing the perceived performance of the system. We leverage this observation with PERSES, a statistical data allocation scheme to reduce the performance impact of reconstruction after disk failure.PERSES reduces degradation from the perspective of the user by clustering data on disks such that data with high probability of co-access is placed on the same device as often as possible. Tracedriven simulations show that, by laying out data with PERSES, we can reduce the perceived time lost due to failure over three years by up to 80% compared to arbitrary allocation.

Reliability challenges for storing exabytes

“…In this manner the failure of one storage brick could be contained to the smallest number of overlying data sets. This is a similar approach to that pursued by the Perses project [18].…”

Can We Group Storage? Statistical Techniques to Identify Predictive Groupings in Storage System Accesses

“…We class an application as grouping for layout when the goal of the grouping is to lay out data on physical media such that colocated data has a high probability of coaccess. We showed instances of this application type when using groups to reduce power consumption [Wildani and Miller 2010] and improve availability [Wildani et al 2014]. In these scenarios, space is relatively plentiful but groups can not change quickly, since the change requires a layout overhead.…”

“…Recent studies indicate that categorical grouping does not have the flexibility necessary for modern, shifting workloads [Wildani et al 2014]. One silver lining of the massive amounts of data that modern systems create is that it is easier to train statistical learning systems on systems with high IOPS, since we have more data to support or contradict any prior calculations.…”

Order Statistics of Inverse Pareto Distribution

“…Naghettini et al (1996) [8] estimated the upper tail of flood peak frequency distribution by the inverse Pareto distribution. For other applications, see Wildani et al [9].…”