Browse Prior Art Database

Management framework for efficient live migration of virtual machines running migration-aware applications Disclosure Number: IPCOM000200260D
Publication Date: 2010-Oct-03
Document File: 4 page(s) / 38K

Publishing Venue

The Prior Art Database


We suggest a mechanism to optimize the execution of live VM migration by leveraging the knowledge about the application running within the VM and about the environment in which the VM is running, as well as by proactively re-configuring the application and the environment, in a lock-step with the progress of the live VM migration process.

This text was extracted from a PDF file.
This is the abbreviated version, containing approximately 30% of the total text.

Page 01 of 4

Live virtual machine (VM) migration is a widely used
technology to move running VMs from one physical machine to
another with minimal application disruption. Although the live
VM migration process is relatively short and non-disruptive,
it does influence both the application running within the VM
(there is a short period of downtime and/or performance
degradation) and the environment in which it is applied (e.g.,
network bandwidth and the resources of the source and
destination physical machines are consumed by the migration
process). In many cases, the live VM migration operation is
rather costly. There are several factors that can be
considered as part of the migration 'cost' concept - e.g.,
completion time, application downtime, performance
degradation, migration resource utilization, etc.

    The problem addressed by this invention is how to
optimize the execution of a given live VM migration so that
its cost would be minimal.

    In some cases it might be possible to reduce some of the
cost factors without penalty in any of the other cost factors
being considered. In other cases, a tradeoff between different
cost factors can be considered (e.g., limiting the network
bandwidth allocated for migrating the VM's state is likely to
increase the downtime during the VM switchover). Moreover, in
different scenarios the importance of each cost factor could
be different (e.g., in case of emergency evacuation of a host
due to expected hardware failure, it might be highly important
to complete the live VM migration as soon as possible, while
paying some performance penalty during the migration process).

    One approach to address this problem is to apply generic
optimization techniques, considering the virtual machine as a
'black box' running an unknown software stack. For example,
data compression techniques could potentially decrease the
bandwidth consumed by the live migration process, as well as
decrease the downtime (if the last portion of the dirty pages
is also compressed and transferred more efficiently). However,
this approach can not leverage internal knowledge of the
application stack, and therefore can not achieve the most
optimal conditions for live VM migration. For example, if the
application stack has some low priority internal processes
which can be suspended for the period of live migration and by
this significantly decrease the rate of dirty pages, the above
generic approach would not be able to leverage such an

    Another approach, which exists in IBM POWER systems, is
to allow the application vendors to develop hooks into the
live VM migration process which would be triggered within the
guest OS at different stages of the live VM migration process
(e.g., before VM is suspended on the source host and after it
is resumed on the target host). This approach does enable
certain level of application-aware optimization (such as the

Page 02 of 4

one mentioned above



but it considers only the local

conditions within this particular VM being migrated, and is
not enoug...