Description

we propose the storage management of online data stores that concurrently support both range queries and dynamic updates . Over in expensive hardware we reduce the data serving latency through higher storage contiguity; improve the performance predictability with limited query-update interference and configurable compaction intensity; and decrease the storage space required for file maintenance through incremental compactions. Our main insight is to keep the data of the memory and disk sorted and partitioned across disjoint key ranges. In contrast to existing methods, when incoming data fills up the available memory of the server, we only flush to disk the range that occupies the most memory space. We store the data of each range in a single file on disk, and split a range to keep bounded the size of the respective file as new data arrives at the server. To achieve the scalable Data without the latency problem we introduced the novel storage layer so that we efficiently manage structured data which is stored in different locations and in different workloads. A storage layer at each server manages the memory and disks to persistently maintain the stored items. Across diverse batch and online applications, the stored data is typically arranged on disk as a dynamic collection of immutable, sorted files. The data is dynamically partitioned across the available servers to handle failures and limit the consumed resources. To a large extent, the actual capacity, functionality and complexity of a data store is determined by the architecture and performance of the constituent servers. Despite the prior indexing research (e.g., in relational databases, text search), data stores suffer from several weaknesses. Periodic compactions in the background may last for hours and interfere with regular query handling leading to latency spikes.