Transactional In-Page Logging for multiversion read consistency and recovery

Abstract

Abstract-Recently, a new buffer and storage management strategy called In-Page Logging (IPL) has been proposed for database systems based on flash memory. Its main objective is to overcome the limitations of flash memory such as erase-beforewrite and asymmetric read/write speeds by storing changes made to a data page in a form of log records without overwriting the data page itself. Since it maintains a series of changes made to a data page separately from the original data page until they are merged, the IPL scheme provides unique opportunities to design light-weight transactional support for database systems. In this paper, we propose the transactional IPL (TIPL) scheme that takes advantage of the IPL log records to support multiversion read consistency and light-weight database recovery. Due to the dual use of IPL log records, namely, for snapshot isolation and fast recovery as well as flash-aware write optimization, TIPL achieves transactional support for flash memory database systems that minimizes the space and time overhead during normal database processing and shortens the database recovery time.

MLC-flash-friendly logging and recovery for databases

“…More recently, research has started exploring recovery for DBMSs over flash-memory storage systems, e.g., [1,11,13,14,18]. Excellent designs are such as a shadow-paging-based FlagCommit scheme to keep track of the transaction status for efficient transaction recovery in SLC-based DBMSs [18] and a transactional in-page logging (TIPL) approach to store updates into the specified log sector within a page for SLCbased databases [13]. Different from the past work, we are interested in the exploring of recovery designs with steal and no-force policies over ARIES-based recovery.…”

“…In the past years, excellent ARIES-based recovery designs were proposed for different purposes, including those for multi-core and multi-socket hardware [8,9], large-scale distributed storage systems [12,20], and multi-thread and multi-server supports [4,6]. More recently, research has started exploring recovery for DBMSs over flash-memory storage systems, e.g., [1,11,13,14,18]. Excellent designs are such as a shadow-paging-based FlagCommit scheme to keep track of the transaction status for efficient transaction recovery in SLC-based DBMSs [18] and a transactional in-page logging (TIPL) approach to store updates into the specified log sector within a page for SLCbased databases [13].…”