I/O not being properly load balanced on EVA

[Andrew Scott_3]

We're running HP-UX 11.31 with an EVA 8100 (active/active controllers), and we've found that ALL I/O is being sent to just one of the EVA controllers.

How do I identify which paths go to the second controller and how do I get the OS to send I/O down those paths?

[Torsten.]

IMHO this is normal, only the owner controller does the I/O via all available pathes.

[Andrew Scott_3]

But aren't these EVAs supposed to have Active/Active controller pairs and the Agile I/O in 11.31 supposed to automatically balance all the traffic among all available paths?

For example:
# scsimgr lun_map -H 64000/0xfa00/0xb

LUN PATH INFORMATION FOR LUN : 64000/0xfa00/0xb

Total number of LUN paths = 8
World Wide Identifier(WWID) = 0x600508b4000b50e80000800000e40000

LUN path : lunpath28
Class = lunpath
Instance = 28
Hardware path = 0/3/0/0/0/0.0x50001fe15010f07e.0x4003000000000000
SCSI transport protocol = fibre_channel
State = ACTIVE
Last Open or Close state = ACTIVE

LUN path : lunpath29
Class = lunpath
Instance = 29
Hardware path = 0/3/0/0/0/0.0x50001fe15010f07f.0x4003000000000000
SCSI transport protocol = fibre_channel
State = ACTIVE
Last Open or Close state = ACTIVE

LUN path : lunpath26
Class = lunpath
Instance = 26
Hardware path = 0/3/0/0/0/0.0x50001fe15010f079.0x4003000000000000
SCSI transport protocol = fibre_channel
State = STANDBY
Last Open or Close state = STANDBY

LUN path : lunpath27
Class = lunpath
Instance = 27
Hardware path = 0/3/0/0/0/0.0x50001fe15010f078.0x4003000000000000
SCSI transport protocol = fibre_channel
State = STANDBY
Last Open or Close state = STANDBY


The active paths correspond to paths to a single controller. If I so much as flip a single standby path to active, ALL the standby paths go Active and the Active paths flip to standby. I though these suckers were supposed to be able to do Active/Active?

[Torsten.]

The EVA does ALUA (Asymmetrical Logical Unit Access). It is balancing the ownership between the controllers (more or less). Accessing the non-owner is much slower.

[Andrew Scott_3]

By ownership do you mean the controller preference set in Command view?

Because we had all luns owned by a single controller. I've gone and flipped half of them to the other controller, but the paths didn't switch.

Why did they all end up on the same controller at creation?

[Torsten.]

See also

http://h30499.www3.hp.com/t5/System-Administration/EVA-on-hp-ux-11-31/m-p/5222168#M467036

[Torsten.]

Let us see

# scsimgr get_info -D /dev/rdisk/disk...

for one of the disks.

[Andrew Scott_3]

# scsimgr get_info -D /dev/rdisk/disk7

STATUS INFORMATION FOR LUN : /dev/rdisk/disk7

Generic Status Information

SCSI services internal state = ONLINE
Device type = Direct_Access
EVPD page 0x83 description code = 1
EVPD page 0x83 description association = 0
EVPD page 0x83 description type = 3
World Wide Identifier (WWID) = 0x600508b40006857a0000f00000bb0000
Serial number = "PBA23D89SUR04J"
Vendor id = "HP "
Product id = "HSV210 "
Product revision = "6220"
Other properties = ""
SPC protocol revision = 2
Open count (includes chr/blk/pass-thru/class) = 1
Raw open count (includes class/pass-thru) = 0
Pass-thru opens = 0
LUN path count = 32
Active LUN paths = 16
Standby LUN paths = 16
Failed LUN paths = 0
Maximum I/O size allowed = 2097152
Preferred I/O size = 2097152
Outstanding I/Os = 10
I/O load balance policy = round_robin
Path fail threshold time period = 0
Transient time period = 120
Tracing buffer size = 1024
LUN Path used when policy is path_lockdown = NA
LUN access type = T10 Asymmetric Active-Active
Asymmetric logical unit access supported = Both implicit and explicit
Asymmetric states supported = ao_sup, an_sup
Preferred paths reported by device = Yes
Preferred LUN paths = 0

Driver esdisk Status Information :

Capacity in number of blocks = 4194304000
Block size in bytes = 512
Number of active IOs = 0
Special properties =
Maximum number of IO retries = 45
IO transfer timeout in secs = 30
FORMAT command timeout in secs = 86400
START UNIT command timeout in secs = 60
Timeout in secs before starting failing IO = 120
IO infinite retries = false