New way of handling RAID 5?
As an Oracle DBA, I have heard and read a lot about RAID 5 and what it can do to database write performance. Now I understand RAID 5 has been improved to the point that all of the things that caused the overhead are now being handled in disk cache, and write times are now close to that of RAID 0+1. Has anyone else heard of this? Has anyone had any experience with it? I have a client who must choose between RAID 0+1 and RAID 5. The decision will have a large bearing on available disk space. Thanks for any input.
Software/Hardware used:
Software/Hardware used:
ASKED:
January 14, 2005 4:37 PM
UPDATED:
August 28, 2008 6:47 PM
Answer Wiki:
This depends on your disk subsystem.
Our Oracle database resides on a SAN attached IBM shark and we have enough write cache in front that we do not have a write penalty for using raid-5.
To see all answers submitted to the Answer Wiki: View Answer History.
Write, or for that matter Read Cache does make a difference to performance. However Raid 5 and Raid 0+1 will always be different ( 0+1 better ). As a DBA I take this view, Raid 5 is ‘cheaper’ than 0+1 ( byte for byte ). As customers want an ever increasing amount of data on-line people start to worry about hardware cost. So, if the data is important RAID 1, if I/O speed is important, add in RAID 0. If cost takes priority over performace and resilience that use RAID 5.
If you are the kind of DBA that it constantly being told to improve performance then stick with RAID 0+1 as ‘you’ know exactly on which disk/spindle the data resides.
If in the future space comes under pressure and you need a stopgap solution, you can always change 0+1 to 5, but not the other way round.
Disk is so cheap now,that you should simply buy lots.
Regards
JJ