IBM Hub/Switch Installation Manual page 443

High performance storage system release 4.5
Table of Contents

Advertisement

7.6.2.7.4
Mirrored Fileset Recovery Speed
Mirrored fileset recovery time must be considered when configuring the system. Mirrored fileset
recovery can be tedious and slow. The DFS fileset is recovered using HPSS metadata, which
requires many SFS accesses.
The three time consuming steps when recovering mirrored filesets are:
Dumping the HPSS fileset information. The rate for this step is approximately 60 entries
per second.
Restoring the fileset to Episode. The rate for this step is approximately 60 entries per
second.
Loading Episode DMAPI handles into the HPSS metadata. This is by far the most time
consuming step, and is highly dependent on SFS speeds. The rate for this step is
approximately 12 entries per second.
The above rates can be used to approximate the recovery times for mirrored filesets. For example,
the recovery time for a mirrored fileset with 100,000 objects is approximately 3.2 hours (1666
seconds for metadata dump + 1666 seconds for Episode fileset recovery + 8333 seconds for HPSS
metadata updates = 11665 seconds = ~3.2 hours).
So, it can be seen that recovery of a lost Episode fileset can take a very long time. If multiple filesets
need to be recovered, many of these steps can be done in parallel, but disk and SFS limitations will
quickly outweigh any amount of parallelism.
7.6.2.7.5
Cloning, Replication, and Fileset Backup
Aggregates managed by the Episode XDSM (DMAPI) do not support cloning. Fileset replication
and movement can not be done because they require cloning. DFS recommends the use of cloning
to backup filesets. These mechanisms cannot be used with HPSS/DFS filesets.
HPSS/DFS filesets that are mirrored do not have to be backed up since they can be restored from
HPSS metadata. The restored mirrored fileset will only be able to restore data that HPSS knows
about. Any file deleted by mistake cannot be recovered on mirrored filesets. Users should be made
aware of this fact and the proper precautions taken.
HPSS/DFS archived filesets must be backed up because not all data is migrated into HPSS and the
name space information is not recoverable from HPSS metadata. The only way to back up archived
filesets is with the DFS utility 'fts dump'. The problem is, the fileset is locked for updates while
the dump is taken. Filesets with a large amount of objects and data may be locked from updates for
a long period of time.
7.6.2.8
HPSS/XFS Usage Guide and Limitations
This section of the document provides information to help system administrators determine the
best way to use the HPSS/XFS interface. Both XFS and HPSS have limitations that will impact
performance when using the HPSS/XFS interface. Some of these limitations are the result of
implementation decisions, others are hardware dependent. This section will address these
limitations, explain why they exist, when and if the limitation will change, and how to configure
your system to minimize the impact of these limitations.
HPSS Installation Guide
Release 4.5, Revision 2
Chapter 7
September 2002
HPSS User Interface Configuration
443

Advertisement

Table of Contents
loading

This manual is also suitable for:

Hpss

Table of Contents