VNX disk storage

[Mark Parsons]

HPUX11.11 and HPUX11.23

Hi,

We have recently purchased VNX disk storage from EMC but there seems to be some discrepencies between what is shown in disk devices in sam.

System 1:

0/10/0/0.4.32.0.0.0.0   4          Unused   --              0   DGC     V ^ ³
³³ 0/10/0/0.4.32.0.0.1.1   4          LVM      vgVNX       25600   DGC     V   ³
³³ 0/10/0/0.4.32.0.0.1.2   4          LVM      vgVNX       25600   DGC     V   ³
³³ 0/10/0/0.4.32.0.0.1.3   4          LVM      vgVNX       25600   DGC     V   ³
³³ 0/10/0/0.4.32.0.0.1.4   4          LVM      vgVNX       25600   DGC     V   ³
³³ 0/10/0/0.4.32.0.0.1.5   4          LVM      vgVNX       25600   DGC     V   ³
³³ 0/10/0/0.4.32.0.0.1.6   4          LVM      vgVNX       25600   DGC     V   ³
³³ 0/10/0/0.4.32.0.0.1.7   4          LVM      vgVNX       25600   DGC     V   ³
³³ 0/10/0/0.4.32.0.0.2.0   4          LVM      vgVNX       25600   DGC     V   ³
³³ 0/10/0/0.4.32.0.0.2.1   4          LVM      vgVNX       25600   DGC     V

So we can see one disk "controller" (0mb) and 9 disks of 25600mb

System 2:

0/2/1/0.4.32.0.0.0.0   4          Unused   --          40960   DGC     VN   ³
³³ 0/2/1/0.4.32.0.0.0.1   4          Unused   --          40960   DGC     VN   ³
³³ 0/2/1/0.4.32.0.0.0.2   4          Unused   --          40960   DGC     VN   ³
³³ 0/2/1/0.4.32.0.0.0.3   4          Unused   --          40960   DGC     VN   ³
³³ 0/2/1/0.4.32.0.0.0.4   4          Unused   --          40960   DGC     VN   ³
³³ 0/2/1/0.4.32.0.0.0.5   4          Unused   --          40960   DGC     VN   ³
³³ 0/2/1/0.4.32.0.0.0.6   4          Unused   --          40960   DGC     VN   ³
³³ 0/2/1/0.4.32.0.0.0.7   4          Unused   --          40960   DGC     VN

So we can see 8 x 40960mb disks but no "controller" is visible (disks have not been allocated into volume groups yet)

System 3:

0/10/0/0.25.2.0.0.0.1   4          LVM      vgomniVN    40960   DGC     V   ³
³³ 0/10/0/0.25.2.0.0.0.3   4          LVM      vgomniVN    40960   DGC     V   ³
³³ 0/12/0/0.24.2.0.0.0.0   4          LVM      vgomniVN    40960   DGC     V   ³
³³ 0/12/0/0.24.2.0.0.0.2   4          LVM      vgomniVN    40960   DGC     V   ³
³³ 0/12/0/0.24.2.0.0.0.4   4          LVM      vgomniVN    40960   DGC     V

So we can see 5 x 40960mb disks but no "controller" is visible

Does the "controller" need to be visible? Which is correct?

The only thing that is different is that System 1 is at HP11.11 whereas System 2 and System 3 are at HP11.23.

(All three systems also have internal disk and eva storage - not shown above)

Any suggestions.

Kind Regards - Mark P.

[Matti_Kurkela]

The SCSI protocol requirement is that LUN 0 must exist, since that is where you send the SCSI Report LUNs command. The LUN 0 can be a regular storage LUN, it does not necessarily have to be a controller device. There is no single "right" way.

But if LUN 0 is a regular storage LUN, you may run into trouble if you later need to unpresent that storage LUN while still using any other LUNs: if LUN 0 does not exist, the only way to detect the other LUNs would be to blindly probe for each of them and wait a while for an answer. This is extremely inefficient, so many OSs (including HP-UX) won't implement it, causing the other LUNs of the same SCSI target to become inaccessible if LUN 0 is unpresented.

Some storage system models solve this by making the LUN 0 a "controller device" which can always be presented to any server, so that you are free to present/unpresent real storage LUNs however you want without needing to think about LUN 0.

All this is applicable to FibreChannel only because the FC implementation of HP-UX 11.23 and older essentially makes FC behave like SCSI. HP-UX 11.31 has a "native" FC implementation which no longer has to apply unnecessary SCSI abstractions to FibreChannel.

(Of course 11.31 can optionally provide old-style /dev/[r]dsk/cXtYdZ device files in addition to the new /dev/[r]disk/disk* style ones; but when you learn the new features of 11.31, you'll soon see that the legacy device files are just a compatibility overlay on top of the new storage subsystem.)

In some EMC products, the presence of controller devices is configurable: you can enable them if you want. The controller devices may also be needed for some advanced/administrative functionality, like SYMCLI access.

[chris huys_4]

Hi Mark,

It shouldnt give problems, having a lun0 or not,  for fiberchannel storage presented to  hp-ux 11.11 or hp-ux 11.23 hosts. Well I cant remember that I ever got into problems with it.

As long as the vnx storage model is listed as supported on this url of hp;

http://www.hp.com/products1/serverconnectivity/mass_storage_devices.html

;everything should work, however it is configured.

And its " ioscan -fn commandline output"  that should be listed of lun output, not sam output of lun output. sam is for configuring users and printers and adding a kernel module/changing a kernel variable. ;)

On HP-UX 11.31, there are some problems , with "online" removing a lun0 after having presented a lun0 to a hp-ux 11.31 host.  But the problems are mostly cosmetic. HP-UX 11.31 , will use the ctl device file, if lun0 is not available, for use to send scsi inquiries to all luns beneath a certain controller, I think. But if for a certain time lun0 is available, which is thus used for scsi inquiries of the other luns, "beneath" that lun0, and then that lun0 get online removed, while the OS could show the other luns in NO_HW, but the luns and the data on the luns ,should  still be available, as the OS will begin sending the scsi inquiries via the ctl device file. A reboot certainly and a scsimgr replace_wwid maybe, will clear the "cosmetic NO_HW" problem.

Best Regards,

Chris