Originally posted by: SystemAdmin

Does anyone have any experience with attaching AIX Pseries systems to Hitachi 9900 storage Arrays?

We are attaching systems to this storage through brocade 4700 directors. FCS cards are 4 GB single port adapters, 5759 adapters.

According to Hitachi we should be connecting pt2pt for the init_link, but our IBM GS contractors are saying that it should be AL, and it will negotiate it. Does anyone have any thoughts on this setting?

Also, with shared storage to 2 AIX serves. We have 2 P595, full partition with Veritas for the VM, and VCS for the HA software.

Again IBM GS says to use single_path for the reserve_policy, I think this is wrong. I no that for concurrent access to these system we should use reserve_policy=no_reserve, however I think we should also use this setting for disks that will be shared and only online on one system at a time.

Any thoughts on the reserve policy would alos be helpful.

Thanks again

John O.
#AIX-Forum

Originally posted by: SystemAdmin

Hi John.
Lets see if I can try to speak as the old AIX dude and not the
IT architect I am today :)=
I'd use a top down approach, have a look at the Veritas documentation,
cause as far as I remember that dictates how some of the reservation
policies are to be set.
They are a bit different on VCS and HACMP as far as I remember.

As for the point to point and auto negotiate, well I'd have a look in the
user guide for the adapter and make up my own mind. Because AFAIR both
HDS and IBM are right :)=

// Jesper
#AIX-Forum

Originally posted by: jvk

John,

for init_link use 'al' and not 'pt2pt' because formal testing is done with the cards set to "al". This includes regression testing and qualification testing.
In this case 'al' tries a point to point connection to a switch, then tries an arbitrated loop connection if that fails. 'pt2pt' tries a point to point connection to a switch, then stops if that fails. So, both setting try a switch connection first, but "al" is called 'al' because it also tries arbitrated loop if the switch connection fails.

Regarding second Q -- do you use MPIO? If yes, then as far as I know there is an agreement between IBM and Hitachi about tested/supported configurations, and in case of MPIO with Hitachi box, one needs to use default MPIO (which is 'fail-over' mode only with 'single path' reservation and with a right ODM disk support supplied by Hitachi. In any case Hitachi dictates and tests how it should be set so I would go to Hitachi for latest info. Be prepared that in case you don't use 'single path', you drop off from IBM support if problem will be related to this.
#AIX-Forum

Originally posted by: jvk

John,

for init_link use 'al' and not 'pt2pt' because formal testing is done with the cards set to "al". This includes regression testing and qualification testing.
In this case 'al' tries a point to point connection to a switch, then tries an arbitrated loop connection if that fails. 'pt2pt' tries a point to point connection to a switch, then stops if that fails. So, both setting try a switch connection first, but "al" is called 'al' because it also tries arbitrated loop if the switch connection fails.

Regarding second Q -- do you use MPIO? If yes, then as far as I know there is an agreement between IBM and Hitachi about tested/supported configurations, and in case of MPIO with Hitachi box, one needs to use default MPIO (which is 'fail-over' mode only with 'single path' reservation and with a right ODM disk support supplied by Hitachi. In any case Hitachi dictates and tests how it should be set so I would go to Hitachi for latest info. Be prepared that in case you don't use 'single path', you drop off from IBM support if problem will be related to this. In any case for shared disks IBM dictates HACMP or GPFS so with Veritas you're more with Veritas support and yes, if disks are accessed from more than one node at time you'll need to set 'no reserve' reserve policy.
#AIX-Forum