Re: Do you still ned RAID 5 if you follow OFA?

I agree. Indeed I see many sites where the system admin has created a very large RAID 0,1 or RAID 5 array then created mount points which were given to the DBA. The DBA blissfully separated members of redo log groups from each other, each control file from the other, datafiles for tables away from datafiles for indexes, unaware that the mount points were all over the disks.

Indeed I have a situation now on a Solaris box where the customer ordered 20 very large disks to RAID. How do I go about specifying the RAID and OFA with that? I can get really good I/O with striping, but I would also like mirroring in places. But I would also like to separate from the datafiles the files (control, redo, archived log files) I would use to recover the datafiles should disaster strike.

OFA has a number of purposes.

First is to have a systematic way of naming and placing files. A DBA may not immediately know where to find a certain type of file, but it would take just a little bit of poking around to find a file on a 22 mount point OFA vs. a 5 mount point OFA.

Second is to assure files which should not be put together are not put together. One control file should go on one device with its own controller, another on another device.

Third is to level out the i/o. Tables should be separated from indexes, which should be separated from rollback segments. This prevents contention and also allows one to RAID based on the type of i/o involved.

The write penalty for RAID 5 is 1/n where n is the number of disks used to create the array. Received on Sun Dec 06 1998 - 11:33:45 CST