English (En)French (Fr)
Home  //  Research Research Results Decreasing Live VM Migration Down-Time Using PDF
Error
  • Unable to load Cache Storage: database
  • Unable to load Cache Storage: database
  • Unable to load Cache Storage: database
  • Unable to load Cache Storage: database
  • Unable to load Cache Storage: database
  • Unable to load Cache Storage: database
  • Unable to load Cache Storage: database
  • Unable to load Cache Storage: database
Decreasing Live Virtual Machine Migration Down-Time Using a Memory Page Selection Based on Memory Change PDF
PDF
Print
E-mail

Fereydoun Farrahi Moghaddam and Mohamed Cheriet, Senior Member, IEEE
Synchromedia Lab, Ecole de technologie supérieure, Montreal.


Published in International Conference on Sensing and Control (ICNSC), pp. 355-359, April 10-12, 2010.
This work is accessible here: http://ieeexplore.ieee.org/stamp/stamp.jsp?arnumber=05461517

Abstract— Seamless migration of Virtual Machines (VMs) will guarantee the Service Level Agreements during the live data center migration which is an important step for allocation and management of Information and Communication Technology (ICT) infrastructures and resources across the globe. At our knowledge, only simplistic models have been used to describe the governing process of memory migration. In this paper, first, we formulate a mathematical model for virtual machine memory transfer. Then, we show the limitations of such a transfer. Finally, we introduce and evaluate a more efficient method for memory transfer.


Virtualization

Virtualization technology gives advantages in three areas: efficiency, control and move-ability. In terms of efficiency, with virtualization technology a data center can save on energy, space, and hardware usage by server consolidation techniques. This technology also gives a high control over virtualized servers by adding the ability to create, delete, and modify a virtual machine server on fly. It also enable making template/snapshot of a server. The last and the most important advantage of virtualization technology is offline/live VM migration. Especially with seamless live VM migration, achieving goals such as dynamic energy efficiency, hardware maintenance without service interruption, and redundancy and security is more and more possible.

A virtual machine

For a VM to migrate, it is necessary that memory is copied completely to the destination before cpu in source is deactivated and cpu in destination is activated.

VM before migration
VM after migration

Memory Migration

Memory is divided to many memory pages. In normal memory migration these memory pages will copy to destination while the cpu is active on source host.

Initial status
Copying all memory pages from source to destination

While memory pages are copying from source to destination some memory pages in source is changed. This is because the cpu in source is active and using the memory. These changed memory pages are called "dirty pages" and are shown in red color in above picture. Next step is to recopy these dirty pages to the destination. This phase is called "warm-up" phase.

Warm-up phase

Warm-up phase will repeat until the number of recopied dirty pages are equal to number of new dirty pages. In this time the cpu in source will be deactivated and the rest of dirty pages will copy to the destination. This phase is called down-time phase because both cpu in source and destination are deactivated.

Down-time phase

After the down-time phase the cpu in destination will be activated and memory and cpu in source will be released.

Final status - Migration completed

Memory Migration Down-Time

In this research we provide a formulation for the amount of memory which could be copied to the destination in warm-up phase as follows:

Amount of copied memory vs time

This formula is correct for the case if dirty pages are created in a uniform way in the memory. If this graph meet the memory size of a virtual machine, then the down-time will be a very small amount of time. If this graph do not meet the memory size there will be a down-time for the virtual machine during migration.

When memory graph do not meet the memory size line

There is a relation between migration down-time and percentage of memory change in the virtual machine:

Memory change rate and down-time relation

If the memory change is uniform, the sequence of selection of dirty pages to migrate does not matter. The simulation result shows this and the accuracy of our memory migration model as follows:

Simulation result for uniform memory change

But this research claim that if the memory change is not uniform, then right selection of dirty pages to migrate; play a big role to decrease the down-time of migration. The idea is to copy the dirty pages with less probability of change first, and then copy the dirty pages with high probability of change. This will help to reduce the number of memory pages which needs to be recopied. Simulation results for different page selections show the efficiency of proposed way of memory migration.

Probability Density Function (PDF) for memory pages of a VM
Simulation result for non-uniform memory change


Conclusion


  • Based on link and uniform memory change properties, seamless VM migration of VM is possible.
  • Page selection based on memory change PDF decrease the down-time of migration and may increase the chance of seamless VM migration for non uniform memory changes.
 

Canarie Logo