Re: Using ADSM to back-up large volumes
2000-01-12 07:50:58
Subject: |
Re: Using ADSM to back-up large volumes |
From: |
Richard Sims <rbs AT BU DOT EDU> |
Date: |
Wed, 12 Jan 2000 07:50:58 -0500 |
Chris - Restorals are the weakness of an "incremental forever"
backup scheme, and as you can imagine has been a
perennial topic of discussion on the List. Here is a reposting of
some approaches I have accumulated over time.
Richard Sims, BU
Restoral performance - Avoid use of the COMPression option
for backups, as the client will have
to uncompress every file being
restored!
- Use Collocation...to the extent that
you can afford it in Backups.
Collocation by FILespace will optimize
restorals but cost a lot in tapes and
tape mount time.
- ADSMv3 Small File Aggregation helps
speed restorals.
- Do Selective backups periodically to
effect a "full" backup and thus a
complete, contiguous image of the
filespace.
- ADSMv3 "No Query Restore" speeds
restorals by eliminating the
preliminary step of the server having
to send the full repertoire of file
objects it has for the client, and the
need for the client to traverse the
list if it already knows what needs to
be restored.
- Maximize your buffer sizes; but watch
out for performance penalty at certain
TCPBufsize sizes (q.v.).
- Minimize your MOUNTRetention value for
the duration of the restoral so as to
avoid a new tape mount having to wait
for a lingering tape to be dismounted
from that drive.
- Consider disabling sync for that file
system for the duration of the
restoral. There is also the public
domain 'fastfs' program for Solaris
systems, to speed restorals through
use of delayed I/O.
- Minimize other server activity during
the restoral period.
- Restoral works by reconstructing the
file system directory structure first.
The directories for many operating
systems reside in the ADSM database
itself; but if yours goes to a storage
pool, make the storage pool disk (as
via DIRMc).
|
|
|