FC lost connection when high I/O load

[Lucy2009_1]

Hi,
We are facing problem on a PA-Risc N class running HP-UX 11.11. When perform high I/O activity to Netapp Storage,the fiber connection will loss. We have confirm Fiber Cable working fine.
Try swap a few brocade port – no help

Connect directly from Server to Netapp Storage – No problem.

Server HP 4000 O/S HPUX 11.11 HBA Card P/N : A5158A

Will this be a problem?
Topology = PRIVATE_LOOP
All other system's HBAs have point-point. How to change this setting?

[root]# fcmsutil /dev/td0
Vendor ID is = 0x00103c
Device ID is = 0x001028
TL Chip Revision No is = 2.3
PCI Sub-system Vendor ID is = 0x00103c
PCI Sub-system ID is = 0x000006
Topology = PRIVATE_LOOP
Local N_Port_id is = 0x000001
Local Loop_id is = 125
N_Port Node World Wide Name = 0x50060b00000f2a31
N_Port Port World Wide Name = 0x50060b00000f2a30
Driver state = ONLINE
Hardware Path is = 0/5/0/0
Number of Assisted IOs = 3364153
Number of Active Login Sessions = 1
Dino Present on Card = NO
Maximum Frame Size = 960
Driver Version = @(#) PATCH_11.11: libtd.a : Jun 28 2002, 11:08:35, PHSS_26799

[BUPA IS · Accepted solution]

Hello,
Your driver is a bit out of date. Please try installing, at the very least, patch PHSS_31326 which gets you to level 11.11.12 which cures and i/o hangs and various other perfomance and reliability issues .
or down load the latest driver from the software depot to get you 11.11.19

Mike

[Lucy2009_1]

Thanks. That's the plan to install patch. Can anyone tell why this HBA came up with PRIVATE_LOOP? SANSurfer to change? or doesn't matter? The HBA connects to Brocade switch and the switch connects to a NetApp filer.

[Lucy2009_1]

Does anyone know how to change "Topology" type?

[BUPA IS]

Hello ,
I do not have any brocade switches .
Generally though the topology is determined by the settings on the primary port on the switch. Most connections from servers to external disk arrays use the point to point method. Loops are often used to provide interconnections between chains of disks inside arrays and the link flows from the switch and is looped in and out of each component.
Adjusting one link in the loop may cause the whole thing to stop. In order to change the topology the entire setup from end to end (and back again if applicable) will have to be analysed (using the switch managment toolset )and all the links adjusted accordingly.
I know it is not a full answer but it might get you started.

Mike

[BUPA IS]

Hello
Sorry after submitted the previous post I noticed in your original question says:

> Connect directly from Server to Netapp Storage â No problem

Does this mean that the brocade switch is not involved in this particular link?
In which case the topology may be set by the filer and may be unchangeable. For the performance issuse, you may also want to check the firmware and patch levels of the netapp filer as well .

I hope this is of some help

Mike

[chris huys_4]

Hi,

Change the domain id of the brocade switch the Nclass hba /dev/td0, is connected to, to something else then 8 and the system will be fine... ;)

Greetz,
Chris
PS. ioscan -fn output, would have shown the problem.

[Lucy2009_1]

Hi Chris,
Thanks for the input. Do you mean to change /dev/td0 to something like /dev/td5 from the switch side? I am not sure how this will help the problem of intermittent i/o lost. Please explain.

[chris huys_4]

Hi,

> Thanks for the input. Do you mean to
> change /dev/td0 to something
> like /dev/td5 from the switch side? I am > not sure how this will help the problem
> of intermittent i/o lost. Please explain.
No, no, no. ;)

A brocade san switch, within a specific fabric, needs to have a unique "domain id" within that fabric.

The domain id, can be a number between, 1 and 239.

If you telnet to a brocade sanswitch and after giving in userid/passwd, issue the command switchshow, you will see what the domain id is set to.

f.e.
# telnet sw03
userid/passwd

sw03:admin> switchshow
switchName: sw03
switchType: 9.2
switchState: Online
switchMode: Native
switchRole: Subordinate
switchDomain: 13
switchId: fffc0d
switchWwn: 10:00:00:60:69:51:31:57
switchBeacon: OFF
Zoning: ON (cl1)
port 0: id N2 Online E-Port 10:00:00:60:69:80:71:da "sw19" (upstream)
port 1: id 2G Online F-Port 50:00:1f:e1:50:04:20:39
port 2: id N2 Online F-Port 50:00:1f:e1:50:04:20:3d
port 3: id N2 Online F-Port 50:00:1f:e1:50:00:94:f9
port 4: id 2G Online F-Port 50:06:0b:00:00:27:0c:a0
port 5: id 2G Online F-Port 50:06:0b:00:00:2c:e9:2e
port 6: id N2 Online F-Port 50:06:0b:00:00:38:b8:8e
port 7: id N2 Online L-Port 1 public
port 8: id N2 Online F-Port 50:06:0b:00:00:10:47:b8
port 9: id N2 Online F-Port 10:00:00:00:c9:2c:cc:ce
port 10: id N2 Online F-Port 50:06:0e:80:04:a4:65:04

>> CONCLUSION -- in the above example the domain id equals to 13 (=switchdomain=domain id)

Do the above switchshow output, for both the sanswitch that connects to the fc hba of youre problem hp-ux host and the sanswitch(es) that connect to the netapp filer and check the domain id of these sanswitches.

One of the domain id will be set to 8. And 8 is a domain id that cant be used with hp-ux, as its reserved for private loop devices. (see http://docs.hp.com/en/J2635-90017/J2635-90017.pdf )

So change with the "configure" command on the sanswitch, the domain id is set to 8, to another unique number in that fabric, .. and then the hp-ux host, should connect the hp-ux host fc hba in pttopt_fabric (fcmsutil /dev/td0 ).

If this will not work, post the relevant switchshow output and ioscan -fn output.

Greetz,
Chris