HP-UX 11.31 and EVA4400

[Filippo_1]

Hi guys.

I'm working in a new environment that has 2 x RX8640 connected to a EVA4400 thought SAN.

Knowing that the EVA4400 is an ACTIVE/ACTIVE storage and HP-UX 11.31 has the native Multipath/LB software, shouldn’t the output for the command "scsimgr lun_map -D /dev/rdisk/disk226" show all the paths as ACTIVE/ACTIVE instead of ACTIVE/STANDBY?


# > scsimgr lun_map -D /dev/rdisk/disk226
LUN PATH INFORMATION FOR LUN : /dev/rdisk/disk226

Total number of LUN paths = 4
World Wide Identifier(WWID) = 0x600508b4000c0bcd0000400001500000

LUN path : lunpath183
Class = lunpath
Instance = 183
Hardware path = 0/0/6/1/0/4/0.0x50014380025d236d.0x402b000000000000
SCSI transport protocol = fibre_channel
State = ACTIVE
Last Open or Close state = ACTIVE

LUN path : lunpath182
Class = lunpath
Instance = 182
Hardware path = 0/0/6/1/0/4/0.0x50014380025d2369.0x402b000000000000
SCSI transport protocol = fibre_channel
State = STANDBY
Last Open or Close state = STANDBY

LUN path : lunpath181
Class = lunpath
Instance = 181
Hardware path = 0/0/14/1/0/4/0.0x50014380025d2368.0x402b000000000000
SCSI transport protocol = fibre_channel
State = STANDBY
Last Open or Close state = STANDBY

LUN path : lunpath180
Class = lunpath
Instance = 180
Hardware path = 0/0/14/1/0/4/0.0x50014380025d236c.0x402b000000000000
SCSI transport protocol = fibre_channel
State = ACTIVE
Last Open or Close state = ACTIVE

In this case, you guys think this might be happening because of the ALUA_ENABLE property that should be set to "FALSE":

# > scsimgr get_attr -D /dev/rdisk/disk226 -a alua_enabled

SCSI ATTRIBUTES FOR LUN : /dev/rdisk/disk226

name = alua_enabled
current = true
default = true
saved =

Thanks in advance.

Filippo

[Steven E. Protter]

Shalom,

HP-UX 11.31 as opposed to earlier versions of the OS has more flexibility and built in multi-pathing.

That is why it shows active/active.

What you say is perfectly normal.

SEP

[Filippo_1]

Hi,

what if I change the ALUA_ENABLE property to FALSE, ins't that better in therms of performance?

One more question, what am I suppose to do in order to get all paths working as ACTIVE?

Thanks,

Filippo

[Filippo_1]

Hi guys, any help on this subject?

Thanks!

[Steven E. Protter]

Shalom again:

You ask:
what if I change the ALUA_ENABLE property to FALSE, ins't that better in therms of performance?

Do you have a performance problem?

Performance problems normally come from how the LUNs are configured and how they are used.

Very commonly, people put write intensive Oracle databases on raid 5 LUNS, when a raid 1 or raid 10 LUN would provide better write performance.

SEP

[Duncan Edmonstone · Accepted solution]

Filippo,

Yes the EVA4400 is active/active, but IO to a non-owning controller is proxied to the owning controller for most IO (for all read IOs which is normally >75% for most workloads).

Some MPIO systems (I think the windows DSM for EVA for instance) will take this into account and just send write IOs to the non-owning controller. The HPUX 11iv3 MPIO subsystem doesn't bother doing this (it doesn't add much perf anyway - IIRC less than 5%).

So what you are sseing is correct for an EVA4400. If you want to get more perf out of an EVA on HP-UX, always use at least 2 LUNs so you can have 1 LUN owned by each controller - you might then do something like LVM extent based striping across the LUNs.

Do not start playing with the ALUA_ENABLE value for the LUN, you'll put yourself in an unsupported configuration.

HTH

Duncan

[Filippo_1]

Hi SEP.

We don't have performance problems. We are using RAID 1 on the EVA.

I was wondering why don't we get all the paths to the device working on ACTIVE state?

And for that reason, my disk acess performance could be better (if were ACTIVE/ACTIVE) than my actual paths configuration (ACTIVE/STANDBY)?

Thanks,

Filippo

[Filippo_1]

Thanks guys.

That was helpful. I am closing the thead.

Points for all.

Filippo

[Filippo_1]

Closing.