Jie Zheng T S Eugene Ng Zhaolei Liu Rice University Kunwadee Sripanidkulchai NECTEC Thailand VEE14 Motivation Live migration can migrate single VM efficiently Existing solutions on migrating VMs in a multitier application can cause serious performance degradation ID: 468958
Download Presentation The PPT/PDF document "COMMA: Coordinating the Migration of Mul..." is the property of its rightful owner. Permission is granted to download and print the materials on this web site for personal, non-commercial use only, and to display it on your personal computer provided you do not modify the materials and that you retain all copyright notices contained in the materials. By downloading content from our website, you accept the terms of this agreement.
Slide1
COMMA: Coordinating the Migration of Multi-tier Applications
Jie
Zheng
, T. S. Eugene Ng,
Zhaolei
Liu
Rice University
Kunwadee
Sripanidkulchai
NECTEC, Thailand
VEE’14Slide2
Motivation
Live migration can migrate single VM efficiently.
Existing solutions on migrating VMs in a multi-tier application can cause serious performance degradation.Slide3
Multi-tier ApplicationSlide4
Split-components ProblemSlide5
Existing SolutionsSlide6
Problem Formulation
Given a set of n VMs in a multi-tier application.
Minimize the performance degradation, or
impact
.
TM[
i,j
]:
communication traffic rates between VMs prior to the start of migration.Slide7
COMMA
“
COordinating
the Migration of Multi-tier Applications”
The first migration coordination system for multiple VMs.
Centralized architecture.
A two-stage scheduling.Slide8
Two-stage Scheduling
First stage: migrates static content.
Second stage : migrate dynamically generated content.Slide9
First Stage
Migrate VMs in parallel and finish all VMs’ data migration at the same time.
The amount of migrated data is fixed.
Adjusts each VM’s migration speed according to its virtual disk sizeSlide10
Second Stage
Migrate dynamically generated content.
After the first stage, a set of dirty blocks are recorded.
The minimum migration speed of a VM must be greater than its block dirty rate.
VMs are divided into groups.Slide11
Inter-group Scheduling
To minimize the impact.
Brute-force
O(2
n
*n!)Slide12
Inter-group Scheduling(Cont.)
To minimize the impact.
Heuristic
Repeatedly take the largest rate element (
rate
,
VM
i
,
VM
j
) from the list L, check if
VM
i
and
VM
j
are already in the group set S.
Merge group or insert new groups.
Find the permutation of groups in S with minimal impact
O(n!)Slide13
Intra-group Scheduling
Starting the VMs’ migration at the same time is not an efficient use of available migration bandwidth.Slide14
Intra-group Scheduling(Cont.)
Estimate the migration time of each VM.
Assume the minimal required speed for each VM is equal to the VM’s maximal dirty rate
“Finish migration at the same time”
Set different start time to each VM.
Extra available bandwidth will be allocated to the VMs to minimize the total migration time of the group. Slide15
Evaluation-Setting
Six physical machines.
Kernel-based Virtual Machine(KVM)
Multi-tier application:
RUBiS
Includes web server, application server, database server.
Performance metric: average response time of the request from clients per second.Slide16
Migrate 3VMs with different strategies.
Evaluation-PerformanceSlide17
Evaluation-Impact
Migrate 4 VMs from at most three physical machines to the other three.Slide18
Evaluation-EC2
Migrate 2 SPECweb2005 VMs from a university campus network to EC2 instances.Slide19
Conclusion
Propose COMMA, the
first coordinated live VM migration system for multi-tier applications.
Based on a two-stage scheduling algorithm to coordinate the migration of VMs.
The goal is to minimize the migration’s impact on inter-component communications.