System Administration
cancel
Showing results for 
Search instead for 
Did you mean: 

HP-UX 11.31 native multi-pathing and EMC Clariion cx-380

 
Deepak Seth_1
Regular Advisor

HP-UX 11.31 native multi-pathing and EMC Clariion cx-380

we are running into some kind of performance issue on one of vpar on rx7640 (oracle 10g) . I have implemented native multipathing of 11.31 by putting the correct paramters on clariion . The clariion is an active/active array as we use flare code 26 . Does it means that all the paths needs to be active. Anybody else using clariion - hpux 11.31 and native multipathing ?
Attacing the list of 2 vpars from the same rx7640 . One of the host is showing one path active and other standby and other hosts shows all path active (that's the one which have ocassional performance issues).



13 REPLIES
Sanjay_6
Honored Contributor

Re: HP-UX 11.31 native multi-pathing and EMC Clariion cx-380

Clarion Arrays when connected and zoned to 11iv3 servers should be configured for FO mode 4 (will show up as SPC protocol revision in the scsimgr get_info -D /dev/rdisk/ output).

It should show up as ACTIVE / STANDBY paths.

check "Active LUN paths" & "Standby LUN paths" in the scsimgr output.

What type of performance issue ? Any other error in the logs.

Hope this helps.
Regards
Deepak Seth_1
Regular Advisor

Re: HP-UX 11.31 native multi-pathing and EMC Clariion cx-380

i configured them correctly . please see the below output . The only thing i noticed is the I/O load balancing policy is set to path_lockdown in this vpar . Other vpar showing as "round robin" . Don't know where that setting comes from.
please see the attached file . The first system is the one having problem and second one is fine.

please see my other attachment(with initial question) for scsimgr lun_map -D output

attached is out of scsimgr get_info -D

Deepak Seth_1
Regular Advisor

Re: HP-UX 11.31 native multi-pathing and EMC Clariion cx-380

here is the output.

Re: HP-UX 11.31 native multi-pathing and EMC Clariion cx-380

Deepak,

It looks very much to me like you _don't_ have the correct settings for that LUN on the Clariion - I'd go and check these 2 LUNs again on the Clariion and make sure you have all the same parameters set.

Once your happy with it, you may also have to recreate the DSFs, so de-activate the VG, and remove the DSFs, then re-run ioscan and insf

HTH

Duncan

HTH

Duncan
Sameer_Nirmal
Honored Contributor

Re: HP-UX 11.31 native multi-pathing and EMC Clariion cx-380

It looks like the host initiator type 10 (HP No Auto Trespass) is not setup on the problematic vpar host.

I would first check if the initiator type is set to 10 for this host on the clarrion array. Also I would confirm this parameter is set "OptionsSupported ArrayType" in the agent.config file.
Deepak Seth_1
Regular Advisor

Re: HP-UX 11.31 native multi-pathing and EMC Clariion cx-380

i checked over and over again and settings are fine on clariion side . attaching is a screen shot from clariion . my agent.conf looks like this and it is same on all the host.

(herhxp02 root):/opt/Navisphere/bin> grep -v "#" agent.config
clarDescr Navisphere Agent
clarContact John Smith, 800-555-1212
device auto auto
poll 60
eventlog 100

other 2 vpars - herhxp01 and p03 are fine and they have same config file as well as similar setting clariion .

the setting were fine some months back and recently changed when i did some zoning changes - made each HBA see multiple sp ports . but i did for others vpar also.
Sameer_Nirmal
Honored Contributor

Re: HP-UX 11.31 native multi-pathing and EMC Clariion cx-380

Ok. I think the hostname needs to be a FQDN.

Try using FQDN for this host. Check agentID.txt and HostIdFile.txt files.
Deepak Seth_1
Regular Advisor

Re: HP-UX 11.31 native multi-pathing and EMC Clariion cx-380

where are these 2 files

agentID.txt and HostIdFile.txt files

I was thinking about the FQN but its not easy to change. I may need to de-register the inititors in clariion . Beside that edit option is just showing the short name of the host. So looks like i probably need to unmount all FS , deactivate the VG and then remove and then re-register the paths on clariion again . but that probably means some downtime and don't know even if after all this it will show the FQN for the host.
Deepak Seth_1
Regular Advisor

Re: HP-UX 11.31 native multi-pathing and EMC Clariion cx-380


the only difference i could figure out so far is in the output of scsimgr get_info -D display under "LUN access type " column ( few more columns as well) .

Host A - The one having Issues
---------------------

LUN Path used when policy is path_lockdown = NA
LUN access type = NA
Asymmetric logical unit access supported = Both implicit and explicit
Asymmetric states supported = NA
Preferred paths reported by device = Yes
Preferred LUN paths = 0

Host B - The one having No Issues
-------------------
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, u_sup
Preferred paths reported by device = Yes
Preferred LUN paths = 0

Any idea about the above differences . What you folks see in your system with similar setup.

Sameer_Nirmal
Honored Contributor

Re: HP-UX 11.31 native multi-pathing and EMC Clariion cx-380

The major difference is "I/O load balance policy" which is "path_lockdown" for the problematic host. I think it should be "round_robin". I think because of the policy "path_lockdown", you see the difference in the other parameters. You need to change the policy.

http://bizsupport1.austin.hp.com/bizsupport/TechSupport/Document.jsp?lang=en&cc=us&taskId=120&prodSeriesId=417289&prodTypeId=18964&objectID=c01685302

Reg to FQDN, maybe the host was registered with short name initially. Since the host can access the LUNs, I think this is not a problem here.
Deepak Seth_1
Regular Advisor

Re: HP-UX 11.31 native multi-pathing and EMC Clariion cx-380

i already changed the policy to roind robin but no change . look at my last reply with some more difference i am seeing .

I don;t know whether this is causing any performance impact or not . We got some ERP session runnig slow but overall not a big issue . However i do see i/o going on all 6 path - sar -L out shows that. strange but i in way its good for the o/s if it goes to both SP controller - real active / active .

thanks for your inputs.
WayneHP
Frequent Advisor

Re: HP-UX 11.31 native multi-pathing and EMC Clariion cx-380

Deepak,

HP-UX is not seeing the LUNS presented correctly on the bad host. You will always get bad performance until you correct this if you send I/O down one of the standby paths. This is what ALUA, Asymmetric Logical Unit Access mean. All paths are not the same. Yes, the array with the HW version is active-active but is is only servicing the LUN on one active controller at a time. The other controller COULD host the LUN and will display a STANDBY path when configured correctly. I also only see you displaying one of the paths set correctly. Here is the version of the doc I have on EMC CLARiiON ALUA. I also have another larger pdf, if you want it please let me know and I will email it.

As a first step to get your performance back up you can use the scsimgr command to send all I/O to a preferred path and test each path one at a time for performance. Here you can get the full bandwidth on one HBA, no HA or Load Balance and you identify the good and bad paths. There should be an equal number of paths each based on your output.

Example:
scsimgr set_attr -D /dev/rdisk/disk0 -a load_bal_policy=preferred_path -a preferred_path=0/3/1/0.0x21000020371972eb.0x0
preferred_path you will get from your scsimgr lun_map -D command test each one. I use the DT tool for this.

You could then set load_balancing to weighted_rr and give the "GOOD" paths a better weight than the â BADâ paths until you fix the issue to get you out of this performance fix. Now you make use of all HBAs but if you have a control crash in the CLARiiON you will have less than idea I/O. If the controller goes down all I/O will still round robin on the left over paths and you will be good, BUT if the failed controller comes back and does not take possession of the LUN(s) again you will have terrible performance.

I think to really fix this you will need to go in on a weekend. Disable the VG using vgchange â a n, unpresent the LUNS to the bad server and use rmsf -x to clear t the stale paths. Halt the server, represent the LUNS using the same setting you used before and boot the server. You may also need to mv /etc/lvmtab to /etc/lvmtab.old then run vgscan â a to recreate.

If all goes well they will come back correctly. I had issues before with the CLARiiON and this is the only way I could get the LUNS back right.
Deepak Seth_1
Regular Advisor

Re: HP-UX 11.31 native multi-pathing and EMC Clariion cx-380

thanks everybody . It turned out be just a simple command to be used to turn on the native multi pathing (ALUA) .

scsimgr set_attr -D /dev/rdisk/disk66 -a alua_enabled=0
#
Validate them using
scsimgr get_attr -D /dev/rdisk/disk66 -a alua_enabled
#

That's fixed the lun_map command showing 3 paths ACTIVE and other 3 as STANDBY.

However that doesn't means the performance issues are resolved . But that's another battle for some other time.

Appreciate everybody suggestions.