Slow RMAN Backups

  • From: Naqi Mirza <naqimirza@xxxxxxxxx>
  • To: oracle-l@xxxxxxxxxxxxx
  • Date: Thu, 17 Aug 2006 23:51:16 -0700 (PDT)


I was wondering whether anyone has experienced poor backup times with oracle 
databases on RAID 5? We are using netbackup in conjunction with rman to backup 
a 1.5TB database that takes about 16 hours to complete. A host of rman 
parameters have been tried - combinations of filesperset and maxopenfiles - 
with no benefit seen. Our database uses RAID 5 spread over about a 100 disks. 
We have noticed that no matter what we do the maximum throughput for the backup 
is only 50mb/s. We are using an LTO3 tape library which can write at 60mb/s per 
tape drive. But no matter how many tape drives we allocate for use (we use 3 
channels - 3 drives) we only get 50mb/s as the total throughput - resulting in 
a backup that is over 16 hours. What happens is, if one channel is allocated 
the throughput for that might be around 50mb/s, when a second channel is 
allocated, the total throughput remains around 50mb/s which is distributed over 
the two channels. Adding a third gives no benefit as the
 50mb/s throughtput is then distributed over all three channels.
I have monitored v$backup_aysnc_io and have noticed that there is for the most 
part poor read rates as seen by the effective_bytes_per_second for 
TYPE='INPUT'. From that I can only infer that the slower the read done by rman 
the slower the backup will be - I do not see any problem with respect to writes 
to the tape drives - they will write as fast as they recieve data. We have seen 
with a filesystem backup we are able to achieve a rate of about 100mb/s - 
although this type of backup is different to an rman backup there is quite a 
difference seen in the rate. Just wondering if there is someway to work out if 
the RAID configuration could be the issue here and whether the diskratio 
parameter might be a viable option to try.

Other related posts: