RHEL 5 - 2 out of 4 HBA's being used (Oracle i/o)

[Mike_Swift]

Greetings!

Our storage would like to  remap half the LUNs that are being used for our RHEL 5.3 server to our two additional HBAs (the server has 4 HBA's, but according to SAN data only 2 are being used, these are Qlogic HBA's).     Our current understanding is that almost all our data is passing through two HBAs.  Our purpose for doing this is to spread this load across all four HBAs that are on that server so that Oracle i/o performance improves. Would this be feasible? Any Ideas??

Thanks

Mike..

[Matti_Kurkela]

What multipathing solution are you using? (dm-multipath or something else? Nothing at all?)

On RHEL 4 & 5, the dm-multipath can only perform round-robin load balancing. It should work well enough on average if all the HBAs of your server are the same speed. On RHEL 6, dm-multipath has other load balancing algorithms available, which might be a better choice if you have different-speed HBAs installed. If you are using another multipathing solution, read its documentation to find what type of load balancing algorithm it uses and which are its advantages & disadvantages.

Are the unused 2 HBAs currently unconfigured, or are they idle despite of being correctly configured and zoned at the SAN side?

Is your storage system capable of active/active operation, or is it an active/passive type?

If it is active/passive and you have a multipathing solution configured, perhaps the additional HBAs are connected to the "spare" controller at the SAN side.

On an active/passive storage, you are supposed to use one controller at a time, and switch to the other only if the primary controller fails. In this case, spreading the load across all connections would cause a lot of trespass events, and the actual performance would be reduced.

And of course, having 4 HBAs at the server side won't help much if the storage system is attached to the SAN fabric by 2 HBAs only: the bottleneck would be at the storage-side connections then.

[Mike_Swift]

Hello Matti,

We use Veritas DMP. The storage is active/active. From the server i see that all 4 HBA's are online.

Thanks

Mike.

[Matti_Kurkela]

The fact that a HBA is online means only that it's physically connected into another FibreChannel port that is active. It does not necessarily tell whether any LUNs are accessible through the connection or not.

Are the 2 quiescent HBAs seeing the same storage LUNs as the active HBAs?

(The "lsscsi" command might be very helpful here: run "yum install lsscsi" or pick the lsscsi RPM from the RHEL 5 installation media if you don't have it installed yet.)

If all the HBAs are seeing the same LUNs but the DMP is only using two, it might be a DMP configuration error, and could be fixable locally by you.

But if the quiescent HBAs aren't seeing any LUNs, you might want to verify the SAN zoning & masking configuration with the SAN administrators first: you should verify that the SAN admins have used the correct WWNs for all the HBAs of this system in the SAN configuration. Collect the WWNs of each HBA and send the list to the SAN admins.