Server: H50, AIX 4.2.1, ADSM 3.1.1.5 (yeah, I know,we 're working on the
upgrade)
Client: NT Server 4.0SP5, ADSM 3.1.0.6 (or .7)
We're doing our-world testing for performance of restores, which is currently
estimated to be unnacceptable (i.e. 2 weeks to bare-metal restore an NT campus
server with 100 Gb of data).
We probably have a fairly brain-dead installation, as it was done by a
consultant 2 years ago who rode off, never to be heard from again, and the
in-house admin retired last year without ever really learning the subtleties of
ADSM. Meanwhile, the number of ADSM clients has probably quadrupled to around
100 from what it originally was in 1998.
We've done a series of tests to simulate the effect of collocation (i.e. an
archive of about 10 Gb, which spanned two tapes). In several variations (GUI vs.
command line, etc), restoring the archive produced a top throughput of 1.2 to
1.5 Gb/hr, which makes the estimated restore time for a full NT machine simply
career-threatening instead of asinine and ridiculous.
We are investigating some network issues, but aren't ready to point a finger
there just yet. We're also looking at the configurations of the NT boxes (i.e.
RAID arrays for disks, how are they cached, etc).
A question that came up for me to follow up on:
When doing a restore in a non-collocated environment, ADSM seems to: a) locate a
tape it needs in the 3494 library, b) mount the tape, c) read whatever it needs,
from a few bytes to a few Gb, d) repeat as necessary until it's done. It never
seems to use more than one 3590 drive simultaneously, even though we have four
and some or all of the others may be idle or even have no tape mounted in them.
Is there an option I missed somewhere that says "Use as many drives as you can
during a restore, to stage one tape while reading from another, or even read
from multiple tapes simultaneously"?
Above and beyond that, is there information out in the big wide world as to what
factors to consider and what parameters to check (end-to-end i.e. ADSM server,
network, and ADSM client) to maximize restore performance, especially in the
case of a large amount of data (10-100 Gb) all at once to a single ADSM client?
Please copy me directly on any answer, as I receive the list digest and it seems
to foul up at least two or three days a week.
Thanks in advance.
Steven
Steven Chaba <chaba AT rge DOT com>
Lead Analyst, ADSM/AIX/HACMP (and other amusing and confusing acronyms) Support
Rochester Gas & Electric Corporation
Rochester, NY
|
