Hello!
When we recently had DR the whole restore process was very slow due to the
order that the files were retrieved from tsm. Many extra tape mounts were
done and it was a very tedious process. A user came up with the following
plan to reduce the number of tape mounts and speed up the recovery process.
Is there another more efficient way of achieving the same goal? This just
seems like a lot of work to get the data back more quickly. Any
suggestions?
The tsm server is on the mainframe at level 5.1.6.2 and most of the servers
that this would be done for are sun solaris with oracle data. Thanks!
dsmadmc `id/password` query volume fromdate=mm/dd/yyyy todate=mm/dd/yyyy
mgmtclass=`the offsite one` > save output to a file1.out
spin through file1.out and get a list of the volumes removing report
headers, footers and all extra info using a unix utility (nawk) >
save output to file2.out
generate one query command for each volume in file2.out > save all of
that output in file3.out
file3.out now contains a listing of what files are on what tape in
same order they are on tape.
At DR, I cut out 50 or so lines containing 5mb files and formatted
the dsmc retrieve commands to test the theory. The run time of the
first command was a couple of minutes (to mount the tape and position
it). All subsequent commands ran in 3 seconds. That time covers
establishing a dsmc connection, TSM figuring out the tape is loaded,
positioning the head, pulling the 5m to unix, and terminating the
session.
I'd like to load that information into an Oracle table since it is
easier to manipulate with SQL.
Generate files from the Oracle table that contain the tsm retrieve
commands and put them on a unix host. There are many ways to
organize the data, haven't thought through it that much. Could be
something like date, server, volume in a directory structure.
Back up those directories to TSM following their creation.
This whole process could be run daily to pick up the previous days
activities. That would keep the stress on the TSM catalog at
reasonable levels. I found out at DR that listing out the contents
of volumes can be time consuming.
All data could be purged from the Oracle tables and unix file system
after something like 70 days. That works for our department. If
other departments use it, more intelligent logic could be used.
At DR, we'd restore the directory structures we need (the ones
created in step7).
We run the appropriate files to restore everything we wanted for each
server.
The affect would be a single pass through each tape for each server
with the retrieve commands pulling the data off in the same order
they are on tape.
Joni Moyer
Systems Programmer
joni.moyer AT highmark DOT com
(717)975-8338
|