5137ch04.fm
Draft Document for Review October 14, 2014 10:19 am
158
IBM Power Systems E870 and E880 Technical Overview and Introduction
group or rootvg. The rootvg can be accessed through a SAN, the same as the data that
partitions use. Alternatively, a VIOS can use storage locally attached to a server, either DASD
devices or SSD drives. For best availability, however accessed, the rootvgs should use
mirrored or RAID drives with redundant access to the devices.
4.5.3 Live Partition Mobility
Live Partition Mobility (LPM) is a technique that allows a partition running on one server to be
migrated dynamically to another server.
This feature can be extremely useful on situation that a system needs to be evacuated for
maintenance but its partitions do not allow for downtime. LPM allows for all the partitions to be
moved while running to another servers so the system can be properly shutdown without
impacts for the applications.
In simplified terms, LPM typically works in an environment where all of the I/O from one
partition is virtualized through PowerVM and VIOS and all partition data is stored in a Storage
Area Network (SAN) accessed by both servers.
To migrate a partition from one server to another, a partition is identified on the new server
and configured to have the same virtual resources as the primary server including access to
the same logical volumes as the primary using the SAN.
When an LPM migration is initiated on a server for a partition, PowerVM in the first system
starts copying the state of memory in the first partition over to a destination partition in
another server through each system PowerVM. This is done across a LAN while the initial
partition continues to run. PowerVM has control of I/O operations through I/O virtualization
and keeps track of memory changes that occur throughout the process.
At some point, when all of the memory state is copied from the primary partition, the primary
partition is paused. PowerVM in the second server takes over control of the shared storage
resources and allows the partition now running in that server to resume processing at the
point where the first server left off.
Thinking in terms of using LPM for hardware repairs, if all of the workloads on a server are
migrated by LPM to other servers, then after all have been migrated, the first server could be
turned off to repair components.
LPM can also be used for doing firmware upgrades or adding additional hardware to a server
when the hardware can not be added concurrently in addition to software maintenance within
individual partitions.
In successful LPM situations, while there may be a short time when applications are not
processing new workload, the applications do not fail or crash and do not need to be
restarted. Roughly speaking then, LPM, allows for planned outages to occur on a server
without suffering downtime that would otherwise be required.
Minimum Configuration
For LPM to work, it is necessary that the system containing a partition to be migrated and the
system being migrated to both have a local LAN connection using a virtualized LAN adapter.
The LAN adapter should be high speed for better migration performance. The LAN used
should be a local network and should be private and have only two uses. The first is for
communication between servers, the second is for communication between partitions on
each server and the HMC for resource monitoring and control functions (RMC.)
