Veritas-bu

[Veritas-bu] oracle restore anomaly

2001-09-13 13:11:02
Subject: [Veritas-bu] oracle restore anomaly
From: Jerald_Iverson AT den.invesco DOT com (Iverson, Jerald S.)
Date: Thu, 13 Sep 2001 11:11:02 -0600
we've recently been testing oracle rman restores and have noticed a problem
with the restore speed.  a stream of data with 5 files of 400 meg each can
restore in less than 10 minutes, or it can take well over an hour.  when we
see that the file sizes are growing extremely slow, we abort the restore,
and then restart it, and it (usually) starts restoring at the much faster
rate.  same tape, same drive, same network, same client, writing to the same
disk, just re-starting the restore job.

due to this problem our dba has researched the oracle rman repository and
found where all the backup data is stored which shows which files are in
which streams and which streams are on which tapes and in what order.  he
can then control exactly what rman is doing, and bypass files that are
already restored when we encounter a slow restore and have to abort and
re-start.  this is on our ncr clients restoring from a sun master/media
server.  we haven't seen this slowness while restoring to our sun oracle
clients, so they require no rman tweaking.  they can basically tell rman to
restore the database to a certain point in time and it works great.

tape speed iostats show it to be around 90% busy when working okay, and 5%
busy when it is on a slow restore.  sar data shows that when working the sys
is at around 55%, and usr and wio are less than 3%. when slow, sys is 75%
and usr and wio are still less than 3%.  when we abort the restore, all drop
to either 0 or 1, with the system being 98% idle, so if it were a competing
process we should see the difference in sys % busy.  about the only process
running is rman.

has anyone experienced this problem?  could it be either rman in a spin loop
and performing slow, or could it be when we restart it we might get on a
different network socket?  i believe it is rman itself that is taking the
one stream of data received from netbackup and splitting the blocks of data
into their correct files. 

we'd like to correct this to allow fast efficient easy restores.  any
help/hints would be appreciated.

netbackup 3.4  patch J0850645
solaris 7
sun 450 master server
sun 4500 media server
ncr mp-ras 3.02 client
oracle 8.0.5

thanks in advance,
jerald iverson

<Prev in Thread] Current Thread [Next in Thread>
  • [Veritas-bu] oracle restore anomaly, Iverson, Jerald S. <=