Redo Logs on RAID-6

We migrated our storage from DMX-4 to VMAX.  I was some what uncomfortable  with one of our main architectural decision of going with RAID-6 because of our millisecond SLA’s. Off course I was wrong and our excellent storage gurus were right. I did lot of research about Oracle installations on RAID-6 especially redo logs. Unfortunately I did not find any. The whole purpose of this article to provide some insight into performance of redo logs on RAID-6. Also the intention of this article is not to explain different types of RAID levels, So I will just to cover some basics before introducing redo logs.

RAID stands for  “redundant array of inexpensive disks”  used to provide highly available disks by storing or striping data across multiple hard drives. Storing data across multiple disks also improves read performance.   I will talk about few important RAID levels.

RAID-0:  Striping with no redundancy. Improved performance but no fault tolerance.

RAID-1: Disk Mirroring i.e.. data is duplicated. Improved read performance with minimal impact to writes.

RAID-10: Combination of RAID-0 and RAID-1.

RAID-5:  Striping and parity information stored in a cyclic/rotating fashion. Some impacts to write operation

RAID-6:  Striping and double parity information stored in a cyclic/rotating fashion. Some impacts to write operation


Now coming to redo logs. Typically Oracle recommends RAID-1 for redo logs. As mentioned earlier, RAID-1 provides increased read performance with minimal impact on writes . However this always comes with additional  cost because of hard disk mirroring.  Most of the time, apart from high concurrency ,  redo log  writes happen to be the major  bottleneck for a tuned database because transaction is considered  committed when change is written to redo log files.

With advancement of storage technologies, RAID-1 may not be necessarily required for improved redo performance. Most of the vendors provide storage cache with  the cache itself  being  mirrored for fault tolerance. This means the modified data is considered  committed   by the storage vendor when it is written to cache. All this is transparent to  Oracle. Therefore  performance gain with RAID-1 may be  insignificant. compared to RAID-5 or RAID-6.  We did not notice any performance impact with RAID-6 even with synchronous SRDF because of storage cache.


Tagged: , , , , , , ,

%d bloggers like this: