|
This paper represents original work by Vincent Cordrey, Doug Freyburger, Jordan Schwartz and Liza Weissler. It is copyright to those authors. The USENIX Association has been granted exclusive rights to publish and disseminate this paper in printed form, electronically, and in audio formats. Original publication of this paper by the USENIX Association is in the "Proceedings of the LISA XIII Conference". This paper may not be published or otherwise disseminated except by the USENIX Association for the duration of the Grant of Rights to Publish. During that time, the Grant of Rights permits this paper to be posted on the Author's web site(s) for electronic access. |
As part of a standard technology refresh program, an obsolete, non Y2K compliant Convex 3240 was replaced and the filesystems on it were rehosted to a new file server. Most of the data was destined for a read-only section of the new storage, while home directories were placed in a filesystem with strictly enforced quotas.
One filesystem had never staged-out files to HSM tapes. This filesystem was evacuated to the read-only "legacy" disk space in one pass using Forward Relocation during the production day.
Home directories were handled differently. Since large numbers of project files had been stored under home directories, most of those files were destined for read-only storage. Only dot files and dot directories were relocated to the new, read-write, home directory storage. Similar to a Reverse Relocation, directories and files appearing in the top level of each user's home directory were pre-populated with reverse links pointing to the old storage. The old storage was set to read-only mode by management request. After this, the users were free to replace those links with actual files; the reverse links had been created to reduce the impact of home directory relocation.
The system was still working, but the media and tape drives were aging and failing. It was also too slow for users to relocate their own data. Due to the extremely large number of files (approximately 997,000) pre-populating the new storage with that number of symbolic links was not practical. Therefore, to minimize user impact, relocation of resident files was started using Forward Relocation.
Shortly after the data movement began, management requested that the old system be placed into a read-only state. This changed the availability state to Semi On-line, and data relocation continued using the Forward Relocation algorithm.
Relocation was paused after the resident files were completed. The Convex was taken Off-line and the primary reference point for the user community became the new "legacy" mount point. Shortly thereafter, the HSM system was rehosted on a physically smaller system.
Following rehosting, the new old server was placed On-line in Read-Only mode. In this optional step, a Hybrid Reverse Linker populated the "legacy" storage with Reverse Relocation links. This permitted users to read copies of their files even before they were relocated, reducing the burden of by hand recovery and relocation.
Data movement resumed using Reverse Relocation by retrieving all files on a particular medium and feeding their names to the relocation worker program. Some repair of damaged tapes was done and files from the repaired tapes were retrieved. Backup tapes were also used to retrieve files whose HSM tapes had degraded beyond usability.
On the new storage (a read-only legacy filesystem),
files older than about 1 year were archived to DLT tape.
Once archived, these files on the new storage were replaced
with symbolic links pointing to the nonexistent object,
archive, so that users browsing the filesystem
would be able to view the names of all files available.