October 30, 2008 - Liuba Shrira

Title: Split Snapshots: A New Approach to Old State Storage

Abstract:
Kurzweil says, computers will enable people to live forever and doctors will be doing backup of your memories by late 2030. This talk is not about that, yet. Instead, the remarkable drop in disk costs makes it possible and attractive to capture past application states and store them for a long time. This opens the possibility that features, such as forecasting and auditing formerly relegated to data warehouses and temporal databases, can become available to everyday applications in general databases. The challenge is how to organize past states so that they are ``not in the way'' and ``always there'' when needed. Split snapshots are a recent approach to organizing past states that is attractive for several reasons. Split snapshots are transactionally consistent. Unmodified database programs can run against them, side by side with programs running against the current state. They can be taken with any frequency, kept indefinitely, or garbage-collected at low cost, a useful feature in long-lived systems. Several new techniques underly split snapshots. The integration with the buffer manager exploits native recovery and concurrency mechanisms to create consistent copy-on-write snapshots without blocking, a new kind of snapshot index provides fast access independently of snapshot age or update workload, and a new kind of snapshot storage organization manages disk adaptively without copying and without disk fragmentation. Measurements of a prototype system indicate that the approach is efficient and scalable, imposing minimal ($4\%$) performance penalty on a storage system, on expected common workloads. (Joint work with Ross Shaull and Hao Xu).