post-copy live migration of virtual machines

1

Post-Copy Live Migration of Virtual Machines

Michael R. Hines, Umesh Deshpande, Kartik GopalanComputer Science, Binghamton University(SUNY)

SIGOPS 09’

2

Outline

• Introduction• Performance Gauge• Design• Evaluation• Conclusion

3

Introduction

• Use post-copy live migration• Reduce total migration time and page faults• Implement Dynamic Self Ballooning(DSB) to

improve performance of both Pre-copy and Post-copy

4

Performance Gauge(1)

• Preparation Time– The time between initiating migration and

transferring the VM’s processor states to the target node

• Downtime– The time that service on migrated VM is unavailable

• Resume Time– The time between resuming the VM’s execution at

the target and the end of migration

5

6

Performance Gauge(2)

• Pages Transferred– the total count of memory pages transferred

• Total Migration Time– the sum of all the above times from start to finish

• Application Degradation– the extent to which migration slows down the

applications running in the VM

7

Design(1) -- Approach

• 1. Stop the VM on source host• 2. Copy processor state to the target host• 3. Resume the VM on target host• 4. Begin fetching memory pages over the

network from the source(Bottleneck)

• How to improve?

8

Design(2)

• Demand Paging– If VM’s memory accessed result in page fault, that

can be serviced by requesting the referenced page over the network from the source node.

– ensures that each page is sent over the network only once

– Page fault will slow down the VM– Long-term residual dependency

9

Design(3)

• Active Push– proactively “push” the VM’s pages from the

source to the target even as the VM continues executing at the target

– ensures that residual dependencies are removed from the source host as quickly as possible

10

Design(4)

• Prepaging– using the faulting addresses as hints to estimate

the spatial locality of the VM’s memory access pattern

– Bubbling algorithm– Reduce page faults and the duration of the

resume phase

11

First , set pivot = 0

0 Max

Pivot

Forward Edge of Bubble

0 Max

Pivot

1 0 Forward Edge of Bubble

0 Max

Pivot

Page Fault ! Set pivot to the fault page

Forward Edge of Bubble Backward Edge of Bubble

12

13

Design(5)

• Dynamic Self-Ballooning (DSB)– Reduce the number of free pages transferred

during migration

14

Design(6)—DSB cont.

• (1) Inflate the balloon:– A kernel-level DSB thread in the VM first allocates as much

free memory as possible and hands those pages over to the hypervisor.

• (2) Detect memory pressure:– Memory pressure indicates that some entity needs to access a

page frame right away.– The DSB process must partially deflate the balloon depending

on the extent of memory pressure.

• (3) Deflate the balloon:– Deflation is the reverse of Step 1

15

Evaluation--environment

• Host– 2.8 GHz multi-core Intel Machine– Gigabit Ethernet Switch– 4 to 16GB memory

• VM– 2 virtual cores– 512MB memory

16

Evaluation(1)—total time

17

Evaluation(2)--downtime

18

Evaluation(3)—pages transferred

19

20

Conclusion & Comment

• Another method for live migration• In some case, it performs better than pre-copy• DSB improve the performance of both pre-

copy and post-copy• It’s still unreliable