Autopath

[Alex_320]

Is autopath supposed to create a pseudo device that represents the two multipathed device? I have one physical disk being accessed by two HBAs. I see two devices in 'ioscan'. In SAM, AP enumerates them into a single device but uses the same device file as the one discovered by the HBA. Is this behavior normal or is Autopath supposed to create a hpap pseudo device in /dev/dsk or /dev?

Autopath software sees two autopath adapters and sees two paths for that single target. do we have to manually 'mknod' an hpap device? Or did I miss something?

SCSI ULM is installed and HP AP is ver 2.00.00 for HPUX 11.0/i.

--alex

[John Poff]

Hi,

We just installed Autopath here on some 11i boxes. It looks like the /sbin/init.d/autopathinit script creates the /dev/hpap device at boot time. I think that device is just something that Autopath uses internally. It looks like Autopath still uses the regular /dev/dsk devices for accessing the drives.

Of course, this answer is based on my minutes and minutes of experience with Autopath, so I could be wrong. :)

JP

[Alex_320]

Hi -

Thanks for the response, I did a quick test, verifying if autopath does, in fact, failover a path. In my test, it did not.

Here's what I have:

/dev/dsk/c1t1d1
/dev/dsk/c2t1d1

Autopath selects /dev/dsk/c1t1d1 as the preferred path. NLB is selected as the default load balancing scheme.

I create a VxFS FS on the /dev/dsk/c1t1d1 device and mount the device on /a. I 'dd if=/dev/zero of=//a/foobar' and run 'sar -d 5 9999' and watch the disk access. All is good, I disable the HBA port and see the I/O stop to c1t1d1 but never see the I/O resume until I re-enable the port. Something is not right, I do see the /dev/hpap device in /dev and in 'lsdev'. I feel like its should create a pseudo device that applications can access, w/out the application be aware of two sd devices.

alex

[John Poff]

Hi,

I'm curious. Why are you using NLB? That is 'No Load Balancing'. We use RR [Round Robin] here. Maybe you could switch it to RR and try your test again?

We've used AutoPath [HP], PowerPath [EMC], and whatever IBM calls their product on the ESS Shark. The only product that creates a pseudo device is the IBM software. HP and EMC just use the regular disk device files, but they intercept the reads and writes and redirect them down different controllers to do the load balancing. The IBM software isn't compatible with MC/ServiceGuard.

JP

[Alex_320]

Hi -

Part of the rationale to have NLB set is to ensure that no I/O is being sent across the redundant fabric. Call is a cost thing or a political thing. In either case, the AP driver should failover, unless its the behavior of AP to require actvity on the device to ensure availability.

I hope this not to the be case, I need to control the "steer" of the traffic into a known and controlled direction.

Alex

[Alex_320]

Hi -

Just for kicks, I did set the LB to RR and I ran the same test. It seems not to work either, when I shut one of the two ports down, AP does not failover traffic to one port, rather, it stops all I/O until the port is back online. I waited roughtly 20 seconds for failover to occur.

Alex

[John Poff]

Is the /dev/dsk/c2t1d1 disk a PV link? Forgive me if I'm asking a dumb question but I haven't messed with SANs and fabric before.

JP

[Alex_320]

Hi -

I don't believe they are PV links, must they be?

alex