Bookmark and Share
RSS

Recent Posts


Getting Hands-On with Live Partition Mobility

April 1, 2008

My first experience with live partition mobility came as an observer. A few months back I went to Austin and saw an LPAR move from one POWER6 570 to another.
 
Since then, I've become a live partition mobility user. I move running LPARs between my two JS22 POWER6 blades, which are connected to the same SAN and able to see the same LUNs. Once I got it set up, I haven't had any problems. It functions with the blades just as I saw it work on the larger systems.
 
Checklist
To get started, I made sure I had the enterprise edition of PowerVM. After loading the virtual I/O (VIO) server onto my blade and logging into the Integrated Virtualization Manager (IVM) GUI, I was able to enter my APV key.
 
I did some other checks during set-up. I made sure my network was set to use a shared ethernet adapter instead of a host ethernet adapter, and I made sure the reserve_lock on my SAN disk was set to no_reserve. (It was originally set to single_path.) If you fail to fix your reserve_lock, you'll need to change it. When searching on "change reserve lock," I found some documentation which included the following:
 
To change the setting with an LPAR running, from your vio server run:

# chdev -dev hdisk7 -attr reserve_policy=no_reserve
 
Some error messages may contain invalid information for the virtual I/O
server environment.
 
Method error (/usr/lib/methods/chgdisk):
    0514-062 Cannot perform the requested function because the specified
    device is busy.

The error makes sense because we've mapped the physical device to a virtual
device.

    $lsmap -vadapter vhost3
    SVSA            Physloc                                      Client Partition ID
    --------------- -------------------------------------------- ------------------
    vhost3          U7998.61X.100BB8A-V1-C17                     0x00000007
 
    VTD                   vtscsi3
    Status                Available
    LUN                   0x8100000000000000
    Backing device        hdisk7
    Physloc               U78A5.001.WIH0A68-P1-C6-T1-W5005076801303022-L6000000000000
 

Shutdown the LPAR that is using the device.

Remove the virtual adapter and its mapping.

$ rmdev -dev vhost3 -recursive
hdisk7 deleted
vhost3 deleted

Change the reserve_lock setting.

# chdev -l hdisk7 -a reserve_lock=no
hdisk7 changed

Let's make sure the setting changed.

$ oem_seup_env
(to leave the restricted shell)

# lsattr -El hdisk7
 
Check for
 
reserve_policy  no_reserve

With the changes verified, the next step is to re-create the virtual device using mkvdev.

Using my IVM GUI, I selected the LPAR that I was going to migrate. (In the drop-down menu, go to "mobility" and select "migrate.") I entered the IP address of the machine I was moving to, followed by the padmin password. Then I selected "validate." This verified that my LPAR was ready to move. Once it passed the tests, I clicked on the "migrate" option, and my LPAR moved to the other blade.   
 
Keep Running
The value of this technology became clear to me on a recent customer call. My customer was conducting maintenance. I had to bring the machine down, and the users suffered an outage. Had the customer been on POWER6, live partition mobility would have been a perfect solution here. I could have moved the running LPAR and then brought down the source machine without affecting the workload that was running.

I'm sure that as more customers deploy POWER6 technology, we'll see live partition mobility become more widely adopted. It's extremely useful technology.

Posted April 1, 2008| Permalink

Post a Comment

Note: Comments are moderated and will not appear until approved

comments powered by Disqus
-->