-
there are 32 processors (2 node rac). The plain file was a file created by makefile - not sure if that makes a difference. It might be cpu intensive - but does that explain poor read rates? i see that the effective_bytes_per_second for type input drop when i add more files to the backup. For example at 3 channels (3 drives) maxopenfiles=1 and filesperset=1 , file abc has a read rate of 25mb/s. However if introuduce an additional channel with the same rman parameters that same file abc that previously has a read rate of 25mb/s drops to about 17mb/s. I am taking these vaues out of v$backup_aysnc_io, where type=INPUT.
This is what pushes me towards an io related issue......what do you think?
-
doesn't sound like local I/O to me.
-
Hi,
I have tested a backup of a single datafiles - both to disk and tape. I notice the following. While running the backup - i monitored v$backup_aysnc_io - specifically effective_bytes_per_second for TYPE=INPUT. I see for a backup to tape the rate achieved for the backup is 26mb/s. I then after the backup is done, before the same datafile backup only this time to disk. The rate achieved here differs by an mb, the backup to disk is about 25mb/s. Almost similar.
My confusion here based on previous posts - is why the io rate achieved from disk (the read portion of the backup) drops when more files are introduced into the backup - via additional channels - Or bundled together. We are writing as good as we are reading. The tapes are supposed to be able to achieve at least about 60-70mb/s with respect to writes. however the tests shown in this post show that a particular datafile is being read at 26/mbs and is also being written at that rate. Could there be some other tests that could be carried out to try and further isolate this problem?
