Good old HW device sharing ... yes it is still possible although I am pretty sure it is rarely used these days.

That's why the description has been removed from the books. Only a fragment (in the NDMP device section) still exists.

Anyway - here is the coarse procedure which I had to verify myself with NW 8.2.1:

  1. Connect the drive(s) to all potential storage nodes.

         Usually you need FC connections because the distances are simply too long.

        That's why EMC ' ... does not support SCSI ...' - they simply do not dare so state that there is the physical limitation.

        Do not foget that such devices must/should use a separate controller.

  2. Ensure that the lib control port is only connected to one host.

  3. Install the NW SN software on all hosts which shall become storage nodes.

        You may configure them as storage nodes now. However, AFAIR by defining a rfemote device such node

        automatically became a SN.

  4. Run 'inquire' on each SN to verify that all devices can be seen on all nodes.

  5. Consider other parameters like reserve/release etc. such also can be configured later.

  6. Before you configure the library let me suggest that you understand hardware sharing.

        You do not even need extra HW - simply use an (A)FTD device.

        Unfortunately, you must assign a "Hardware id" (whatever string) to the device and this edit field is grayed out in

        the GUI. But this is not true - you can still set it from the command line ;-) .

        Here is an example to a locally (NW server) attached FTD - please note that each '\' for a Windows device must

        be repeated:

nsradmin> create type: nsr device; media type: file;name:"rd=server_name:D:\\Z";hardware id:1

                        type: nsr device;

                        name: "server_name:D:\\Z";

                  media type: file;

                 hardware id: 1;

Create? y

created resource id 38.0.52.11.0.0.0.0.20.144.253.84.10.54.87.39(1)

nsradmin> create type: nsr device; media type: file;name:"D:\\Z";hardware id:1

                        type: nsr device;

                        name: "D:\\Z";

                  media type: file;

                 hardware id: 1;

Create? y

created resource id 39.0.52.11.0.0.0.0.20.144.253.84.10.54.87.39(1)

nsradmin> q

        Now label and mount a media in either device - the access for the other one will automatically be blocked.

        That's what you will see in the GUI:

           - Monitoring window                           : You will only see the 'active' device instance

           - non-active Device - Information tab    : "hw_id:shared drive in use"

       Of course you can now write and read the volume as usual.

       Next, unmount the media and mount it onto the other device (DO NOT LABEL IT).

       Then verify the changed scenario. That's all about DDS ('dynamic' drive sharing). The only difference is that NW

       will assign a jukebox device really as needed (dynamically).

  7. For the jukebox configuration I am pretty sure that you must use the command line (jbconfig) as the autodetect and

      configuration utilities still cannot read the admin's mind about hardware sharing. (but wasn't this in development :-)) )

      Just answer all questions thoroughly (especially the ones about the hardware id).

      This is it.

Finally you need to configure the clients for the proper storage node list .... or install NW 8.2.1 and use the new feature 'Autoselect torage node'.

Good luck.

I check you number 6 suggestion - yes, this work as you write, ver thank you for explanations. The things became more sensible.

I'm ready to check this with my test lib, but I did not get how I can present my lib controller only for one server - controller path use the same FC link as some of drives. So if I present this path only for one server and isolate from another , another can not see all four drives.

Mmmm...I play with my lib and still have no luck. I did like you say - manually create new drive device (like "rd=my.host.local:/dev/nst0" - as inquire show the drive) and add this drive to selection creteria in pool settings, but when I start the backup job, the job is wait for media. And I have no idea how to put the tape in drive automatically.

Please explain me what I miss in configuration?

If you only have configured the tape drive(s), there is no automatism. This was only for demnstration.

Of course you must finally configure the Jukebox as well.

The library control port may/will be visible from other hosts.

However, you must only use it on one host - where you run jbconfig.

NW will automatically do the rest and send the appropriate commands to the host which controls the library.

BTW - be careful with the pool selectrion criteria I recommend not to use it.

The potential issue is that you limit your pool unnecessarily. And this can cause further trouble.

Often people think that one MUST select a drive to be able to use the pool. This is simply wrong.

The best proof is the pool 'Default' by itself:

  - it has no device selected

  - you are not even allowed to modify it

  - but you can still label a media and write to it

Have you selected this drive in the pool configuration for the pool that is assigned to the backup group to which this client belongs to.

May be it's to complicated for me and I should go play with Rubik's Cube

At first I remove manually added drive. Then I try to add drive to pool selection creteria, - no tape in drive. I try to remove drive from pool selection (and remove all another checkbox) - no tape in drive.

I run jbconfig on storagenode2 - "no scsi tape library found or allways configured". I see in devices folder in NMC 4 devices for storagenode1 (like rd=storagenode1.my.local:/dev/nst0-3), 4 devices for storagenode2 (like "rd=storagenode2.my.local:/dev/tape/by-id/-...-nst), and in StorageNode folder in NMC on storagenode1 - jukebox with drives, on storagenode2 - no devices at all.

All 4 drives is visible by SAN from both host:

cat /proc/scsi/scsi |grep HP

  Vendor: HP       Model: Ultrium 4-SCSI   Rev: H6FW

  Vendor: HP       Model: MSL G3 Series    Rev: 1130

  Vendor: HP       Model: Ultrium 4-SCSI   Rev: H6FW

  Vendor: HP       Model: LOGICAL VOLUME   Rev: 3.00

  Vendor: HP       Model: P410i            Rev: 3.00

  Vendor: HP       Model: Ultrium 4-SCSI   Rev: H6FW

  Vendor: HP       Model: Ultrium 4-SCSI   Rev: H6FW

So I cannot able to add jukebox at storagenode2 neiter by jbconfig, nor by NMC.

p.s. can it be the license issue - I have an "volume" license?

NW's command 'inquire' or "inquire -l" is the one you want to use to verify whether a node sees a device.

You must execute it locally on each host.

It scans all potential SCSI addresses. If it finds a device, it will display the appropriate connection information and a vendor specific string. So on at least one node you should see an entry like this for your MSL (sorry, Windows):

C:\>inquire

scsidev@0.0.0:hp      DVD A  DS8A5LH  1HE3|CD-ROM, \\.\CdRom0

scsidev@2.0.0:COMPAQ  SCSI COMMUNICATECIS2|Unknown Device Type,

      ....

scsidev@18.0.0:GRAU    ITL-5000l       1.7 |Autochanger (Jukebox),

                                           S/N: 0000010007

                                           ATNN=GRAU    ITL-5000l       0000010007

                                           VBNN=00D080FFFF563412

      ....

4 Disk, 2 Tape, 1 CD-ROM, 1 Autochanger (Jukebox), 2 Unknown Device Type, Total: 10

C:\>

As long as this is not visible, NWwill not install/configure it.

You do not have a license problem.

I am pretty sure you have a driver's problem.

It is important that NW uses ...

  - for the tape drive OS- or vendor specific drivers

  - for the library its own drivers.

    So make sure that no other driver blocks the jukebox access.

Ok, I fix it.

As prerequisite  I clean all experimental librarys, drives and storage nodes (and pool selection creteria).

1) go to "main" storage node storagenode1.

2) run

jbconfig -s nwserver.my.local

3) on

step 1 select "2) Configure an Autodetected SCSI Jukebox."

step 2 enter name for jukebox

step 3 select "Autocleaning yes"

step 4 "Bypass auto-configure? (yes / no) [no] yes" <--------- select yes!

step 5 NDMP - no

step 6 "Is ... more than one path defined? (yes / no) [no] yes" <------ select yes!

step 7

Drive  1, element 1, system device name = /dev/nst0,

      local bus, target, lun value = 0.0.0,

      WWNN=500xxxxxxxxx1D4F model LTO Ultrium-4

Device path 1 ? [storagenode1.my.local:/dev/nst0] (enter)

Device path 2 ? storagenode2.my.local:/dev/nst1   <-------- yes, the names is different on one and another servers.

Device path 3 ? (enter)

repeat step6 for other drives.

step 8 finish config.

4) On NMC I see one new lib with name from jbconfig, and 2 new storagenode - storagenode1 with jukebox and storagenode2 without any devices. In librarys I see a new tape library with 4 drives and 8 devices, two devices for every one drive.

The other things is easy - I add a 3 storagenodes to my client on storagenode 2:

storagenode2

storagenode1

nsrserver

So when I start backup task in tape lib I see on one drive one device is in play, and other is like "sharing device in use".

While I try different configurations I spent a lot of time to repair "serial number mismatch" or "The drive serial number has changed" error. After a long investigation I see that the drive with serial number xxxx1D4F on one server have path /dev/nst0, and on other server have path /dev/nst1, and with other 3 drive exactly the same situation exist.

I'm little bit tired for fix udev to change the path for storagenode2, so i recreate by jbconfig second paths for all drives like storagenode1.my.local:/dev/nst0  first path, storagenode2.my.local:/dev/nst1 second.

Thanks for your feedback - good to see that it still works as I have learned it years ago.