Instant recovery for data center savings

摘要

Today's data centers routinely employ triple redundancy, i.e., each disk page of a database or of a key value store is stored three times (or even more, e.g., in database and file system backups). In contrast, write-ahead logging can reduce the cost of database operations and of data centers, assuming suitable techniques for logging, log archiving, backing up, and recovery. The present paper summarizes our work to-date on single-page repair, instant restart, and instant restore; it describes our techniques for self-repairing indexes, single-pass restore, and virtual backups; and outlines the opportunities for single-copy databases, for avoidance of ever taking a backup (full or incremental), yet for availability and reliability matching today's double-or triple-redundant data centers.