Issues with load balance of tape devices

[Prokopets]

Hi!

   I have a problem with load balancing of tape drives that are presented from SAN. During backup, the backup software uses only one fabric (as VTL shows) to write data to tapes. I tried to investigate the situation and found that netbackup is using the "multi-pathed" device instead of "raw" one:

======================

# ./tpconfig -d
Id  DriveName           Type   Residence
      Drive Path                                                       Status
****************************************************************************
0   hcart2039            hcart2 TLD(1)  DRIVE=2
      /dev/rtape/tape16_BESTnb                                         UP
1   hcart2040            hcart2 TLD(1)  DRIVE=1
      /dev/rtape/tape18_BESTnb                                         UP
2   hcart2037            hcart2 TLD(0)  DRIVE=1
      /dev/rtape/tape26_BESTnb                                         UP
3   hcart2035            hcart2 TLD(0)  DRIVE=2
      /dev/rtape/tape27_BESTnb                                         UP
4   hcart2029            hcart2 TLD(0)  DRIVE=5
      /dev/rtape/tape28_BESTnb                                         UP
5   hcart2031            hcart2 TLD(0)  DRIVE=4

..............

# ioscan -km dsf
Persistent DSF           Legacy DSF(s)
========================================

...........

/dev/rtape/tape26_BESTb   /dev/rmt/c70t0d1BESTb
                         /dev/rmt/c71t0d1BESTb
/dev/rtape/tape26_BESTnb   /dev/rmt/c70t0d1BESTnb
                         /dev/rmt/c71t0d1BESTnb
/dev/rtape/tape27_BEST   /dev/rmt/c70t0d2BEST
                         /dev/rmt/c71t0d2BEST
/dev/rtape/tape27_BESTn   /dev/rmt/c70t0d2BESTn
                         /dev/rmt/c71t0d2BESTn
/dev/rtape/tape27_BESTb   /dev/rmt/c70t0d2BESTb
                         /dev/rmt/c71t0d2BESTb
/dev/rtape/tape27_BESTnb   /dev/rmt/c70t0d2BESTnb
                         /dev/rmt/c71t0d2BESTnb

............

======================

As scsimgr shows, the tape device uses path_lockdown load balancing policy

======================

# scsimgr get_info -H 64000/0xfa00/0xbc

        STATUS INFORMATION FOR LUN : 64000/0xfa00/0xbc

Generic Status Information

SCSI services internal state                  = UNOPEN
Device type                                   = Sequential_Access
EVPD page 0x83 description code               = 2
EVPD page 0x83 description association        = 0
EVPD page 0x83 description type               = 1
World Wide Identifier (WWID)                  = "HP      Ultrium 5-SCSI  VD23YLBG001082"
Serial number                                 = "VD23YLBG001082"
..........
I/O load balance policy                       = path_lockdown
..........

======================

and i cannot change this. As i understand, all the tapes are accesible via only one fabric by default. My question is: how (if it's possible at all) can i change the active paths to tape device? Ideally i want to change the active path on the half of viewable tapes to distribute the load to both interfaces.

Best regards,

Philipp.

[VK2COT · Accepted solution]

path_lockdown

Directs I/O requests through a single hardware path. The chosen hardware path is the one that is least ]

loaded when the device is opened. This algorithm is the only one supported for serial devices such as tape

drives, changers, and SCSI controller devices, and does not apply to block devices such as disks.

The "HP-UX 11i v3 Native Multi-Pathing for Mass Storage", 2011 edition:

Policies Supported for Tapes, Changer and Pass-Through Devices
Because of the serial nature of the I/O flow to tape and changer devices, only one path selection mechanism

is supported by estape and eschgr drivers: path lock down (path_lockdown). When opening a tape or

changer LUN DSF, the lunpath onto which all the I/O operations are sent can be selected by the mass

storage subsystem or set by the administrator. By default, the mass storage subsystem internally selects the

optimal lunpath.

The administrator can manually set the lunpath with the lpt_to_lockdown attribute. The

lpt_to_lockdown attribute must only be set when the device is in the UNOPEN state.

Devices claimed by the pass-through driver (esctl) also use path lockdown as the I/O load balance policy.

Cheers

[Prokopets]

Thanks, that worked fine!