any proper settings for scsimgr lun_paths ALUA hpux11.31 EMC Vnx5100 array, database-random access.

[Steve Post]

I ran scsimgr_esdisk and found a lot of parameters.

I know I need ALUA on.

I expected some document from EMC or HP that would tell me what to set these to or risk crashes.

So I'll ask you guys.

People use computers?  HP computers? 

People sometimes hook up disk arrays from EMC to them?  With a fibre connection?

People seem to be using hpux 11.3.

Maybe people use itanium.

So there has to be like maybe 5 people out there?

When you 5 people of Earth set up your EMC array, did you set scsimgr parameters?

Did you live with the defaults?

scsimgr get_attr -D /dev/rdisk/disk5 -a  congest_max_retries  90
scsimgr get_attr -D /dev/rdisk/disk5 -a  max_retries          45
scsimgr get_attr -D /dev/rdisk/disk5 -a  path_fail_secs      120
scsimgr get_attr -D /dev/rdisk/disk5 -a  esd_secs             30
scsimgr get_attr -D /dev/rdisk/disk5 -a  load_bal_policy   round_robin
scsimgr get_attr -D /dev/rdisk/disk5 -a  preferred_path        nothing here.
scsimgr get_attr -D /dev/rdisk/disk5 -a  alua_enabled        true
scsimgr get_attr -D /dev/rdisk/disk5 -a  max_retries          45
scsimgr get_attr -D /dev/rdisk/disk5 -a  max_q_depth        128 (It was 8).

[Torsten.]

I have no idea if this array is ALUA (like HP P6000) or not (like HP P9500), butgoogle findings point to ALUA.

Usually the server is aware of this behaviour, you can check like this:

# scsimgr get_info -D /dev/rdisk/disk54

...

LUN access type = T10 Asymmetric Active-Active

Asymmetric logical unit access supported = Both implicit and explicit

...

[Duncan Edmonstone]

The VNX is the replacement for the EMC Clariion right?

So on the Clariion, as long as you had the failover modes set correctly for HP-UX11iv3 (11.31), then the HP-UX IO stack would auto-sense the ALUA settings and set all parameters appropriately. I know when it was Clariion, you needed to be on failover mode 4. For the VNX I would assume it is similar or the same. This manula here ceratinly gives that impression:

https://elabnavigator.emc.com/emcpubs/elab/config_guide/HP-UX.pdf

See p186

For all the rest of those options, typically I see them left at their defaults except occasionally people change the max_q_depth, but as to how you go about setting that, well there's a whole technical paper on the best way to manage that:

http://bizsupport2.austin.hp.com/bc/docs/support/SupportManual/c02054623/c02054623.pdf

[Matti_Kurkela]

We have EMC Symmetrix arrays, and lived happily with the defaults.

In general, unless the EMC Connectivity Guide for HP-UX and/or the EMC support matrix specifically says you need to change some value, the defaults should be safe.

Having said that, the Appendix D in the latest version of the HP-UX Connectivity Guide is titled "Excessive Path Failovers in LVM PVLink VNX Series and CLARiiON Configurations". It says nothing about the outcome, but I guess excessive path failovers would not cause crashes, but rather would murder your I/O performance. The instructions are rather specific and too long to copy here: you should get the Connectivity Guide from powerlink.emc.com and read it.

[Duncan Edmonstone]

MK,

On 11.31, I think its unlikely that Steve will be using PVLinks - of course the manual isn't clear about whether the same issue can be seen with the 11.31 native MPIO stack. I'd suggest Steve ask EMC for clarification on that, although I suspect their response will be "Use PowerPath"!

[Steve Post]

Thank you all.   I didn't expect an answer.  I merely hoped for one. 

Most of your suggestions I already did.  But I am VERY glad to hear suggestions.   It means I'm not the only person on Earth using this. And the suggestions were all good.  I do not want to put a solved tag on any of them because the responses are all good.

I have been in contact with EMC support. This morning he confirmed I should not change the defaults.  I need the failover set to 4 for ALUA (which I already did).  And the max_q_depth should not be greater than 16.

After things simmer down a bit, I will probably update this thread.   If I find my critical key that fixes my performance problem I will definitely update the thread.  

steve

[Abhzi]

Hi Steve!

Did you get a fix for your performance issue. It'll be helpfu  if you can share!