Last modified 13 years ago Last modified on 04/22/08 15:35:33

Tape Backend handled in dCache

On Write

Files written to dCache, and configured to be stored on the backend Storage System, are collected on write pools and are sorted according to attributes which may derive from their location in the filesystem tree or via a srm space. These attributes are basically mapped to File Families or Storage Sets or however a set of tapes is called in the connected HSM system.

In order to keep the number of mount requests low, the collected files are writting to tape (resp. flushed into the HSM system) in bunches, writing only those, having the same attributes (resp. going to the same set of tapes)

The time, the flush is started can be configured and takes into account the maximum number of files or bytes collected or the time the oldest file is waiting on disk to be flushed.

On Read

If a file, not available on disk, is requested for reading it is scheduled to be fetched from the backend HSM system. The actual read request is halted until the file could be fully restored to disk. There is no mechanism in dCache yet, to collect read requests to optimize tape accesses.

Special error code:

71 : disable request

Future Plans

In the current implementation of the HSM access module, each storage pool (file system) individually decides on the time it starts the bulk transfer into the tertiary system. With the increasing number of pools and due to the limited capabilty of some HSM systems to handle unlimited chaotic store/retrieve requests, we are planing for a module, centrally managing store and retrieve HSM operations. See here for a preliminary design idea.

Created by Patrick