Speaker: Dexter "Kim" Kimball

Title: Using Unison to Provide "Hot Spare" RW volumes

Abstract:

Unison, an open source project (GPL) from Dr. Benjamin Pierce at the University of Pennsylvania Department of Computer and Information Science, can be used to provide fail over for critical RW volumes.

Unison is used to synchronize the on-line "active" volume with an on-line NRTM (Near Real Time Mirror) volume of a different name and mounted at a configurable location.  If the active volume becomes unavailable the NRTM volume can be quickly substituted.

Unison is idempotent and resilient.

It supports manual and automatic, unidirectional or bidirectional synchronization, with or without user interaction, and can merge file differences using external programs.

Unison can also be used to synchronize files to/from laptops or other supported OAFS client devices.

This talk discusses features of Unison and configuration for use with providing "hot spare" RW OpenAFS volumes.

Failover to a "hot spare" RW volume is considerably faster in most circumstances than converting a RO volume to a RW volume, using vos dump/restore, or recovering volumes from a backup system.

Synchronizing the active volume with the NRTM volume does not involve making the RW volume busy during vos release clone operations, which has proved to be critical in a degraded OpenAFS environment that caused some RW volumes to be busy (unwritable) for tens of minutes during vos release clone creation.

Synchronization times are of course proportional to the number of files and the number of changed files, and to the frequency of Unison invocations.

Relative to using other RW volume recovery strategies, Unison typically results in a smaller "loss window," defined as the discrepancy between original and failover volumes due to execution/synchronization time.

Volumes containing mount points are supported (indefinite recursion is avoided.)

Failover methods will be discussed.

Results of tests, benchmarks, and examples of real world production experience (good, bad, and Oops) will be included.

The presenter, Dexter "Kim" Kimball, has been working with AFS since 1993, starting as a technical trainer at Transarc and continuing as an independent consultant for Transarc/IBM, for United Airlines, and currently for the NASA Jet Propulsion Laboratory as an OpenAFS system administrator and developer, where he practices the use of Unison synchronization described in this talk.
 

Slides: PDF PPT