Multipath VERY slow? hp-ux rp3440 and ibm ds4300

[byau]

Hi there-
Have an hp-ux 11iv3 rp3440 with AB379-60101 dual port HBA hooked up to a HP Storageworks 4/8 San Switch hooked up to an IBM ds4300.

Basically the multipath is not working at all. I usually do quick tests of writing files

# dd if=/dev/zero of=1gigfile bs=1024 count=1000000

which takes anywhere between 15 to 30 seconds. In this case it takes over 10 minutes. When I pull the cable from the left port or the right port, it takes the 15 to 30 seconds. When both cables are plugged in it takes over 10 minutes

any ideas? Googled around a bit and tried a few things. Updated the driver to

FibrChanl-01 B.11.31.0909 FibrChnl;HW=A6826A,A9782A,A9784A,AB378A/B,AB379A/B,AB465A,AD193A,AD194A,AD300A

Also found that commonio was not installed on the system so downloaded and installed that.

CommonIO B.11.31.0909 Common IO Drivers

Still makes no difference.

Any ideas? I'm really not sure where else to look right now. I have googled around a bit to try and find how to view or change the configuration of the multipath and have come up blank.


I am not using pvlinks (that I know of). I'm guessing that the native multipathing with the driver is being used. When I ran the ioscan it already showed what I believe is the multipath to the disk. Here is the ioscan output:

# ioscan -f -n -N -C disk
Class I H/W Path Driver S/W State H/W Type Description
===================================================================
disk 3 64000/0xfa00/0x0 esdisk CLAIMED DEVICE HP 73.4GST373455LC /dev/disk/disk3 /dev/rdisk/disk3
disk 4 64000/0xfa00/0x1 esdisk CLAIMED DEVICE HP 73.4GST373455LC /dev/disk/disk4 /dev/rdisk/disk4
disk 5 64000/0xfa00/0x2 esdisk CLAIMED DEVICE TEAC DV-28E-C /dev/disk/disk5 /dev/rdisk/disk5
disk 46 64000/0xfa00/0xc esdisk CLAIMED DEVICE IBM 1722-600 /dev/disk/disk46 /dev/rdisk/disk46

# ioscan -f -n -C disk
Class I H/W Path Driver S/W State H/W Type Description
=======================================================================
disk 2 0/0/2/0.0.0.0 sdisk CLAIMED DEVICE TEAC DV-28E-C /dev/dsk/c0t0d0 /dev/rdsk/c0t0d0
disk 0 0/1/1/0.0.0 sdisk CLAIMED DEVICE HP 73.4GST373455LC /dev/dsk/c2t0d0 /dev/rdsk/c2t0d0
disk 1 0/1/1/1.2.0 sdisk CLAIMED DEVICE HP 73.4GST373455LC /dev/dsk/c3t2d0 /dev/rdsk/c3t2d0
disk 6 0/3/1/0.1.7.0.0.0.0 sdisk CLAIMED DEVICE IBM 1722-600 /dev/dsk/c9t0d0 /dev/rdsk/c9t0d0
disk 30 0/3/1/1.1.7.0.0.0.0 sdisk CLAIMED DEVICE IBM 1722-600 /dev/dsk/c8t0d0 /dev/rdsk/c8t0d0
#

I created the disk using:
# mkdir /dev/vg10
# mknod /dev/vg10/group c 64 0x010000
# vgcreate -s 4 -e 65535 -p 5 /dev/vg10 /dev/disk/disk46
# lvcreate -L 15000 /dev/vg10
(I will eventually be creating three luns with this volume group, the first lun is 15gig)
# newfs -F vxfs /dev/vg10/rlvol1
# mount /dev/vg10/rlvol1 /mnt


Any ideas?

Thanks in advance!

[byau]

more info:
# scsimgr lun_map -D /dev/rdisk/disk46

LUN PATH INFORMATION FOR LUN : /dev/rdisk/disk46

Total number of LUN paths = 2
World Wide Identifier(WWID) = 0x600a0b80001980ae000053504ac4f6ff

LUN path : lunpath35
Class = lunpath
Instance = 35
Hardware path = 0/3/1/1.0x200600a0b81984d5.0x4000000000000000
SCSI transport protocol = fibre_channel
State = ACTIVE
Last Open or Close state = ACTIVE

LUN path : lunpath36
Class = lunpath
Instance = 36
Hardware path = 0/3/1/0.0x200700a0b81984d6.0x4000000000000000
SCSI transport protocol = fibre_channel
State = ACTIVE
Last Open or Close state = ACTIVE

#

Should I change to active/standby?

[kunalsahoo]

Yes....you should use act/pass..also i would suggest to use pvlinks or any hpux multipathing soft ware to do load balancing...

[byau]

is there a document you can point me to on how to use pvlinks? I had assumed that there was native multipathing available and wasn't sure if I needed to use pvlinks

I ended up changing load_bal_policy=preferred path and preferred_path= one of the lun paths and it seems to be working now (still active/active) . So I believe this fixed the problem - am I missing anything?

[Kenan Erdey]

Hi,

ibm ds series is active/passive array. so you must choose one path to be prefferred. you can install ibm tools from dvd and run SMdevices command to see which is preffered path. if you don't choose the right one when os tries to sent data, controller changes the active one repeatly. this is the cause of the problem.

[TTr]

Since the array is active/passive, switching to active/standby should be sufficient. PVlinks would not do anything differently. Instead of using disk46, pvlinks would use both disks c8t0d0 and c9t0d0, one would be primary and the other would be alternate paths.

EXTREMELY IMPORTANT: The active disk device (and path) should be the one that connects to the DS4300 controller that owns the LUN. With active/active setup the array is going crazy switching LUN ownership between its controllers. That's why it takes 10 minutes for your dd test.

Now since the array is indeed active/passive you can NOT do load balancing as mentioned above without some additional steps.

The array is actually "dual active" meaning that it is active/passive but both controllers can do active i/o.

Right now you are using half of the i/o potential of the array and of the server. The reason is that you have created only ONE LUN.
You must create an EVEN number of LUNs.
On the array you assign half of them to one array controller and the other half of the LUNs to the other controller.
Now on the server each LUN will have dual paths. You use the LUN path that corresponds to the array controller that own the LUN as the primary (or active). The other path becomes the standby.

Now you have manual load balancing. Half of your LUNs do i/o one one controller/HBA and the other half of the LUNs do i/o on the other controller/HBA. You have to ensure that the i/o is sperad out to the LUNs evenly so that the load balancing is as close to 50/50 as possible.

[byau]

That's all great info. Thank you!