ASM_PREFERRED_READ_FAILURE_GROUPS

Question

Hi,
the ASM_PREFERRED_READ_FAILURE_GROUPS parameters is very useful for exented/streched Cluster to minise the traffic between all the extende Node.
In the same idea and to design extended cluster with Exadata somebody have any idea of parameter inside ASM to force the mirrorng disk affinity to ensure the replication between site.
The idea is to have multiple failure group inside a local and a replicaton between site to design a veritable stretched cluster.
Any idea ?
Thank a lot.
Edited by: Florent on 4 janv. 2012 10:21

Answer 1

We just got our hands on 1TB flash.  To mirror we would need 2TB.  We have read about ASM preferred read failure groups and how they allow you to have flash as the primary read area and regular HD as the failure group if something were to happen to the flash.  That way you use the 1TB flash and the mirror is the regular HD,
Hi,
Preferred read failure was primarily designed to force ASM read from Device with less latency avoiding "link" contention, very common in RAC Extended.
IMHO make sense use two failgroup (FLASH and Non-Flash) and use "preferred read" feature since you are mixing different technologies
You just must guarantee that WRITE on Non-Flash tech is acceptable because it will be your "bottleneck".
You can take advantage choosing a FLASH failgroup a prefered read. It will improve reading data to your environment.
If you don't  use "prefered read" Instance will read data from both array with different tech.
So, suppose you are having a high load and excessive WRITE, if ASM read data from Non-Flash it will generate much more WAIT Event compared when ASM read from Flash device.
The reason is obvious:
Flash disk have no moving (mechanical) components. This distinguishes them from traditional electromechanical magnetic disks HDD, which contain spinning disks and movable read/write heads.