Efficient Method Of Predicting Storage Device Failure In Datacenter Using Machine Learning Techniques

Abstract

newline vii newlineABSTRACT newlineThe data centers are interconnection of network elements and servers to newlinecompute and store the huge data in the disk storages from the real world and when the newlinedata is required by the external world it is fetched from disk by various business logic newlineand frameworks for the real time operations. These hard disks are maintained in data newlinecenter with the help of top of rack (TOR) switches and servers. The data center newlineframework may use MPLS cloud or the backhaul architecture to store and fetch the newlinedata. Through these networks the data is stored and fetched from the disk storages newlinewhich are maintained in the data centers. The major failures can be seen when the newlinedisk is not reachable or the disk is in repair, sometimes the issue is seen in the MPLS newlinecloud and data backhaul network. However, these network issues are taken care by newlinethe backup path and node recovery mechanisms. Major cost involves when the failure newlineis seen in disk drive of the data center, such disk failures must be identified earlier, newlineand actions should be taken accordingly to avoid service disruption and to reduce the newlinemaintenance cost. Such prediction of failure in disk storage of data center is carried newlineout in this research with the help of Machine learning techniques. Disk manufacturers newlinehave manufactured disk (both HDD and SDD) with SMART parameters as registers newlineembedded in disks. These SMART parameters are measured and monitored by using newlineexternal applications, perform training of learning model by using machine learning newlinetechniques and disk failures are predicted effectively in ahead of time.