SlideShare a Scribd company logo
1 of 38
Live VM Migration
Hwanju Kim
1
Outline
• Live VM migration
• Use cases
• Live migration mechanisms
• Pre-copy live migration
• Post-copy live migration
• Related research
• Energy savings of idle desktops using virtualization
• LiteGreen
• Jettison
• Cloud Micro-Elasticity via VM State Coloring
• Kaleidoscope
2/38
LIVE VM MIGRATION
3
Live VM Migration
• Live VM relocation
• Lively synchronizing memory contents including CPU
states
• Storage is shared in LAN (e.g., NAS)
Network Storage
Configuration Data
VM
Memory ContentMemory Sync
VM
User
4/38
What is “Live”?
• Migration metrics
• Total migration time
• Time elapsed until all VM states including CPU and memory
are transferred
• Load is changed (balanced) after this time
• Downtime
• Time elapsed while a VM is being stopped
• Service is unavailable during downtime
• What is live migration?
• Migration with near-zero downtime
5/38
How to Migrate a VM
• How to synchronize memory contents?
• Stop-and-copy
• Stop the source VM
• Copy its memory contents over network
• Start the destination VM
• Pre-copy
• Copy memory contents over network
• Keep copying only dirty pages iteratively
• Stop the source VM if # of dirty pages is under threshold
• Copy remaining dirty pages
• Start the destination VM
• Post-copy
• Stop the source VM
• Copy CPU states and page tables over network
• Start the destination VM
• Copy its memory contents on demand
Downtime ∝ Memory size
Not live!
Near-zero downtime
Live!
6/38
Pre-copy vs. Post-copy
• Pros and cons
Pre-copy migration Post-copy migration
Eager copy of source VM’s memory Lazy copy of source VM’s memory
- Longer and unpredictable downtime
depending on writable working set
+ Shorter downtime
+ Shorter total migration time - Longer total migration time
+ High performance after migration
- Low performance after migration
due to network page fault
- Waste network bandwidth by pages that
will not touched by a destination VM
+ Effective use of network bandwidth
7/38
Pre-copy vs. Post-copy
• Trade-off
Totalmigrationtime
Downtime
Post-copy
Stop-and-copy
Pre-copy
Live
Since pre-copy live migration is good for
downtime and migration time, it has
been used in most VMMs
• Overhead after migration can be
effectively reduced by prefetching
• Suitable for VM forking and microsleep
8/38
Pre-copy Live Migration
• “Live migration of Virtual Machines [NSDI’05]”
9/38
Post-copy Live Migration
• “Post-Copy Live Migration of Virtual Machines
[VEE’09]”
• Main issue: How to reduce runtime overheads after
post-copy migration
Prepaging (prefetching) policy
Bubbling with single pivot
Bubbling with multiple pivots
10/38
ENERGY SAVINGS OF IDLE DESKTOPS
USING VIRTUALIZATION
Related research
11
Introduction
• How serious is desktop energy consumption?
Source: Greener PCs for the enterprises
12/38
Introduction
• Why nontrivial for desktop energy savings?
VS
Users don’t want ongoing jobs
to be disrupted even when away
Great savings when away
Roughly 60% of office desktop
PCs are left on continuously 13/38
Naïve Method
• Sleep
• ACPI S3 and S4 states
• S3 – standby (suspend to RAM)
• S4 – hibernate (suspend to disk)
• Pros.
• Significant energy savings
• Cons.
• Losing network presence
I expect the torrent
download to
complete after
drinking!
So, don’t sleep!!!!
How to save energy with
handling user’s ongoing or potential tasks 14/38
Existing Methods
• Proxy-based Approach
• WoL (Wake-On-Lan) proxy
• Same subnet, known MAC addresses, manual operation
• Protocol proxy [NSDI‘09, USENIX’10]
• Triggered by a filtered subset of the incoming traffic
• Listening network ports, user input
• Explicit specification before sleep
• Application proxy [NSDI‘09]
• Application-specific stubs
• Complexity for creating each application stub
15/38
LiteGreen Project (Mircosoft)
• LiteGreen: Saving Energy in Networked Desktops
Using Virtualization [USENIX’10]
• Achieving the conflicting goals
• Energy saving and continuous computing
• Eliminating complexity from protocol- or application-
specific approaches
• locating a desktop in local desktop
• for good user experiences
• consolidating idle desktops in a server
• for energy savings
VM Live
migration!!
16/38
LiteGreen Overview
• Architecture
17/38
How LiteGreen Works
• Operations
Hypervisor
stub
VM
Hypervisor
controller
VM VM VM
LiteGreen Server
Desktop
RDP Client
Live migration makes a desktop “always on”
18/38
LiteGreen Demo
• http://www.youtube.com/watch?v=uHnCiRpfRSs
19/38
Problems of Full VM Migration
• Excessive network bandwidth for migration
• VM memory size + alpha (dirty block copies)
• e.g., about 4.27GB for 4GB VM
• “Boot storm” (after lunch)
• Long migration time
• Delayed sleep
• e.g., 38sec for 1VM, 253sec for 8 VMs
• Less energy savings
• Full VM migration after ballooning  ballooning requires
considerable time and I/O
• Consolidation aborted by short idle time
• Long resume time
• Poor user experience
20/38
Jettison
• Jettison: Efficient Idle Desktop Consolidation
with Partial VM Migration [EuroSys’12]
• Goals
• Quick resume
• Good user experience
• Conservation of the network resources
• Efficiency and scalability
• Cost effective
• Reduction in TCO by energy savings
• Idea
• “Partial VM migration” with fetching required parts
on demand
21/38
Partial VM Migration
• Jettison
Hypervisor
stub
Hypervisor
controller
VM VM VM
Jettison Server
Desktop
VM
Sleep
(S3)
Wake-on-LAN
VM
1. Idleness detection
2. Consolidation
4. On-demand fetch
3. Microsleep
5. Reintegration
Procedure
22/38
State Prefetch
• Prefetch for increasing inter-arrivals of remote
faults
• Hoarding
• Based on fetched frame sequence of a previous migration
• On-demand prefetch
• Based on spatial locality
23/38
State Prefetch
• Trace-driven offline analysis
• Page access traces from a user VM consolidated 58
times
On-demand prefetch works well with 20 page window
24/38
Budget Analysis
• Full vs. Parital VM migration
• Assuming 16GiB memory SunFire X2250
• USD 6099
• Full VM migration
• 33.95 USD / desktop / year
• 33.95 x 4 VMs x 3 years = USD 407.40
• Partial VM migration
• 37.35 USD / desktop / year
• 37.35 x 98 VMs x 3 years = USD 10,980.90
25/38
KALEIDOSCOPE: CLOUD MICRO-
ELASTICITY VIA VM STATE COLORING
Related Research
26
Elasticity of Clouds
• Ideal elasticity: Pay-per-use model
• Achieves both QoS and efficient resource utilization
Source: http://astadiaemea.wordpress.com/2010/06/
27/38
What Matters for Elasticity?
• Granularity
• A unit of service delivery and billing
• VM as a unit
• IaaS (e.g., Amazon EC2)
• Coarse granularity
• A VM booting from scratch
• QoS
• Well-known trade-off against resource utilization
• Conservative elasticity
• High QoS, but inefficient resource utilization
• Aggressive elasticity
• Low QoS, but efficient resource utilization
Ideal Cloud!
How about
QoS?
Too slow for
ideal elasticity
28/38
QoS in Clouds
• Dynamic adjustment of worker VM pool
• Amazon EC2
• Auto Scaling
• Elastic Load Balancing
• Load balancing using elasticity
• Load > TH
• Inflate VM pool by requesting additional VMs
• Load < TL
• Deflate VM pool by returning unnecessary VMs
• High threshold
• Achieves aggressive elasticity for efficient resource utilization
• Requires fast VM instantiation for QoS
29/38
Elasticity Needs
• AT&T’s hosting in January 2010
Needs for elasticity
Short-lived workers
30/38
Problems of Current Clouds
• Slow VM instantiation
• Average 2min to boot a VM (Amazon EC2)
• Very fluctuating latencies
• Cold status of new VMs
• Initially empty OS caches
• Performance degradation during peak load
• Inefficient resource utilization of new VMs
• Full memory allocation during short-lived VMs that
require smaller working set
31/38
Micro-Elasticity
• Goals
• Fast VM instantiation
• VM cloning: SnowFlock [Eurosys’09]
• Efficient memory utilization for short-lived VM
• On-demand resource allocation
• Warm status of new VMs
• Prefetching related data: VM state coloring
Color-based fractional VM cloning
32/38
Live VM Cloning
• Trade-off between cloning techniques
Post-copy cloning Pre-copy cloning
SnowFlock [EuroSys’09] Like live migration
Lazy copy of parent’s memory Eager copy of parent’s memory
Short cloning time
Long and unpredictable
cloning time
Low performance after cloning
due to the cold status
High performance after cloning
due to the warm status
Effective use of network
bandwidth
& Possibility of memory
savings
Waste of memory and network
bandwidth by pages that will
not touched by clone VMs
33/38
VM State Coloring
• Effective VM memory prefetching scheme
• Assuming that locality exist within a related region
• Partitioning VM memory into semantically related
regions
• Methods
• Architecture-based coloring
• Introspective coloring
VM memory
=
Uniform binary
state
VM state coloring
34/38
VM State Coloring
• Color map example
• SPECweb Support workload
• Interspersing of different colors in the physical
memory space of the VM
Yello –page cache
Light blue – user data
Dark blue – kernel data
Light red – user code
Dark red – kernel code
Black - free
35/38
VM State Coloring
• Benefits of per-color prefetching against color-
blind prefetching
• Accuracy
• Fewer wasted fetches of unneeded pages
• Efficiency
• Less page faults
• Per-color prefetch tuning
36/38
Implications for Clouds
• QoS and resource use
• Kaleidoscope with TH=90% outperforms Elastic
Clouds with TH=50%
37/38
Summary
• Live migration is a key technique of virtualization
• Pre-copy live migration
• Working well for general workloads
• No performance degradation after migration
• Used by most VMMs
• Post-copy live migration
• On-demand migration
• Efficient bandwidth usage
• Strong for write-intensive workloads
• Assisted by prefetching
38/38

More Related Content

What's hot

Hardware virtualization basic
Hardware virtualization basicHardware virtualization basic
Hardware virtualization basicSanoj Kumar
 
Server virtualization
Server virtualizationServer virtualization
Server virtualizationofsorganizer
 
Virtualization in Cloud Computing
Virtualization in Cloud ComputingVirtualization in Cloud Computing
Virtualization in Cloud ComputingPyingkodi Maran
 
Load balancing in cloud computing.pptx
Load balancing in cloud computing.pptxLoad balancing in cloud computing.pptx
Load balancing in cloud computing.pptxHitesh Mohapatra
 
Network Virtualization Architectural & Technological aspects
Network Virtualization Architectural & Technological aspectsNetwork Virtualization Architectural & Technological aspects
Network Virtualization Architectural & Technological aspectsdeshpandeamrut
 
Introduction to Virtualization
Introduction to VirtualizationIntroduction to Virtualization
Introduction to Virtualizationelliando dias
 
Load Balancing Algorithms - Which one to choose?
Load Balancing Algorithms - Which one to choose?Load Balancing Algorithms - Which one to choose?
Load Balancing Algorithms - Which one to choose?Aditya Kamat
 
Storage Area Network(SAN)
Storage Area Network(SAN)Storage Area Network(SAN)
Storage Area Network(SAN)Krishna Kahar
 
VMware vSphere Storage Enhancements
VMware vSphere Storage EnhancementsVMware vSphere Storage Enhancements
VMware vSphere Storage EnhancementsAnne Achleman
 
Virtualization and its Types
Virtualization and its TypesVirtualization and its Types
Virtualization and its TypesHTS Hosting
 
Introduction to Virtualization
Introduction to VirtualizationIntroduction to Virtualization
Introduction to VirtualizationRahul Hada
 

What's hot (20)

Hardware virtualization basic
Hardware virtualization basicHardware virtualization basic
Hardware virtualization basic
 
Virtualization.ppt
Virtualization.pptVirtualization.ppt
Virtualization.ppt
 
Server virtualization
Server virtualizationServer virtualization
Server virtualization
 
What is Virtualization
What is VirtualizationWhat is Virtualization
What is Virtualization
 
Virtualization in Cloud Computing
Virtualization in Cloud ComputingVirtualization in Cloud Computing
Virtualization in Cloud Computing
 
Load balancing in cloud computing.pptx
Load balancing in cloud computing.pptxLoad balancing in cloud computing.pptx
Load balancing in cloud computing.pptx
 
Network Virtualization
Network Virtualization Network Virtualization
Network Virtualization
 
Network Virtualization Architectural & Technological aspects
Network Virtualization Architectural & Technological aspectsNetwork Virtualization Architectural & Technological aspects
Network Virtualization Architectural & Technological aspects
 
Introduction to Virtualization
Introduction to VirtualizationIntroduction to Virtualization
Introduction to Virtualization
 
Cloud sim
Cloud simCloud sim
Cloud sim
 
1 virtualization
1 virtualization1 virtualization
1 virtualization
 
Virtual machine
Virtual machineVirtual machine
Virtual machine
 
Virtualization security and threat
Virtualization security and threatVirtualization security and threat
Virtualization security and threat
 
Load Balancing Algorithms - Which one to choose?
Load Balancing Algorithms - Which one to choose?Load Balancing Algorithms - Which one to choose?
Load Balancing Algorithms - Which one to choose?
 
Virtualization
VirtualizationVirtualization
Virtualization
 
Storage Area Network(SAN)
Storage Area Network(SAN)Storage Area Network(SAN)
Storage Area Network(SAN)
 
VMware vSphere Storage Enhancements
VMware vSphere Storage EnhancementsVMware vSphere Storage Enhancements
VMware vSphere Storage Enhancements
 
Live VM Migration
Live VM MigrationLive VM Migration
Live VM Migration
 
Virtualization and its Types
Virtualization and its TypesVirtualization and its Types
Virtualization and its Types
 
Introduction to Virtualization
Introduction to VirtualizationIntroduction to Virtualization
Introduction to Virtualization
 

Viewers also liked

Virtualization Security
Virtualization SecurityVirtualization Security
Virtualization Securitysyrinxtech
 
Security and Virtualization in the Data Center
Security and Virtualization in the Data CenterSecurity and Virtualization in the Data Center
Security and Virtualization in the Data CenterCisco Canada
 
Challenges in Cloud Computing – VM Migration
Challenges in Cloud Computing – VM MigrationChallenges in Cloud Computing – VM Migration
Challenges in Cloud Computing – VM MigrationSarmad Makhdoom
 
Hypervisor Security - OpenStack Summit Hong Kong
Hypervisor Security - OpenStack Summit Hong KongHypervisor Security - OpenStack Summit Hong Kong
Hypervisor Security - OpenStack Summit Hong KongRobert Clark
 
Virtualization: Security and IT Audit Perspectives
Virtualization: Security and IT Audit PerspectivesVirtualization: Security and IT Audit Perspectives
Virtualization: Security and IT Audit PerspectivesJason Chan
 
Virtualization 101: Everything You Need To Know To Get Started With VMware
Virtualization 101: Everything You Need To Know To Get Started With VMwareVirtualization 101: Everything You Need To Know To Get Started With VMware
Virtualization 101: Everything You Need To Know To Get Started With VMwareDatapath Consulting
 

Viewers also liked (6)

Virtualization Security
Virtualization SecurityVirtualization Security
Virtualization Security
 
Security and Virtualization in the Data Center
Security and Virtualization in the Data CenterSecurity and Virtualization in the Data Center
Security and Virtualization in the Data Center
 
Challenges in Cloud Computing – VM Migration
Challenges in Cloud Computing – VM MigrationChallenges in Cloud Computing – VM Migration
Challenges in Cloud Computing – VM Migration
 
Hypervisor Security - OpenStack Summit Hong Kong
Hypervisor Security - OpenStack Summit Hong KongHypervisor Security - OpenStack Summit Hong Kong
Hypervisor Security - OpenStack Summit Hong Kong
 
Virtualization: Security and IT Audit Perspectives
Virtualization: Security and IT Audit PerspectivesVirtualization: Security and IT Audit Perspectives
Virtualization: Security and IT Audit Perspectives
 
Virtualization 101: Everything You Need To Know To Get Started With VMware
Virtualization 101: Everything You Need To Know To Get Started With VMwareVirtualization 101: Everything You Need To Know To Get Started With VMware
Virtualization 101: Everything You Need To Know To Get Started With VMware
 

Similar to 6. Live VM migration

5. IO virtualization
5. IO virtualization5. IO virtualization
5. IO virtualizationHwanju Kim
 
Enhanced Live Migration for Intensive Memory Loads
Enhanced Live Migration for Intensive Memory LoadsEnhanced Live Migration for Intensive Memory Loads
Enhanced Live Migration for Intensive Memory LoadsSamsung Open Source Group
 
Latest (storage IO) patterns for cloud-native applications
Latest (storage IO) patterns for cloud-native applications Latest (storage IO) patterns for cloud-native applications
Latest (storage IO) patterns for cloud-native applications OpenEBS
 
DevNexus 2024: Just-In-Time Compilation as a Service for cloud-native Java mi...
DevNexus 2024: Just-In-Time Compilation as a Service for cloud-native Java mi...DevNexus 2024: Just-In-Time Compilation as a Service for cloud-native Java mi...
DevNexus 2024: Just-In-Time Compilation as a Service for cloud-native Java mi...RichHagarty
 
JITServerTalk JCON World 2023.pdf
JITServerTalk JCON World 2023.pdfJITServerTalk JCON World 2023.pdf
JITServerTalk JCON World 2023.pdfRichHagarty
 
ClickOS_EE80777777777777777777777777777.pptx
ClickOS_EE80777777777777777777777777777.pptxClickOS_EE80777777777777777777777777777.pptx
ClickOS_EE80777777777777777777777777777.pptxBiHongPhc
 
CloudOpen Japan - Controlling the cost of your first cloud
CloudOpen Japan - Controlling the cost of your first cloudCloudOpen Japan - Controlling the cost of your first cloud
CloudOpen Japan - Controlling the cost of your first cloudTim Mackey
 
JITServerTalk.pdf
JITServerTalk.pdfJITServerTalk.pdf
JITServerTalk.pdfRichHagarty
 
Virtualization for competitive advantage - Eric Vanderburg
Virtualization for competitive advantage - Eric VanderburgVirtualization for competitive advantage - Eric Vanderburg
Virtualization for competitive advantage - Eric VanderburgEric Vanderburg
 
Software Defined Agility for IBM FlashSystem V9000
Software Defined Agility for IBM FlashSystem V9000Software Defined Agility for IBM FlashSystem V9000
Software Defined Agility for IBM FlashSystem V9000Catalogic Software
 
JITServerTalk Nebraska 2023.pdf
JITServerTalk Nebraska 2023.pdfJITServerTalk Nebraska 2023.pdf
JITServerTalk Nebraska 2023.pdfRichHagarty
 
Cloud Computing: Safe Haven from the Data Deluge? AGBT 2011
Cloud Computing: Safe Haven from the Data Deluge? AGBT 2011Cloud Computing: Safe Haven from the Data Deluge? AGBT 2011
Cloud Computing: Safe Haven from the Data Deluge? AGBT 2011Toby Bloom
 
Presentation citrix cloud platform for infrastructure as a service
Presentation   citrix cloud platform for infrastructure as a servicePresentation   citrix cloud platform for infrastructure as a service
Presentation citrix cloud platform for infrastructure as a servicexKinAnx
 
Cloud stack overview
Cloud stack overviewCloud stack overview
Cloud stack overviewhowie YU
 
(CMP402) Amazon EC2 Instances Deep Dive
(CMP402) Amazon EC2 Instances Deep Dive(CMP402) Amazon EC2 Instances Deep Dive
(CMP402) Amazon EC2 Instances Deep DiveAmazon Web Services
 
1 virtual machines
1 virtual machines1 virtual machines
1 virtual machinesLen Bass
 
20191201 kubernetes managed weblogic revival - part 1
20191201 kubernetes managed weblogic revival - part 120191201 kubernetes managed weblogic revival - part 1
20191201 kubernetes managed weblogic revival - part 1makker_nl
 
Geographically Distributed Multi-Master MySQL Clusters
Geographically Distributed Multi-Master MySQL ClustersGeographically Distributed Multi-Master MySQL Clusters
Geographically Distributed Multi-Master MySQL ClustersContinuent
 
The impact of cloud NSBCon NY by Yves Goeleven
The impact of cloud NSBCon NY by Yves GoelevenThe impact of cloud NSBCon NY by Yves Goeleven
The impact of cloud NSBCon NY by Yves GoelevenParticular Software
 

Similar to 6. Live VM migration (20)

5. IO virtualization
5. IO virtualization5. IO virtualization
5. IO virtualization
 
virtual machine.ppt
virtual machine.pptvirtual machine.ppt
virtual machine.ppt
 
Enhanced Live Migration for Intensive Memory Loads
Enhanced Live Migration for Intensive Memory LoadsEnhanced Live Migration for Intensive Memory Loads
Enhanced Live Migration for Intensive Memory Loads
 
Latest (storage IO) patterns for cloud-native applications
Latest (storage IO) patterns for cloud-native applications Latest (storage IO) patterns for cloud-native applications
Latest (storage IO) patterns for cloud-native applications
 
DevNexus 2024: Just-In-Time Compilation as a Service for cloud-native Java mi...
DevNexus 2024: Just-In-Time Compilation as a Service for cloud-native Java mi...DevNexus 2024: Just-In-Time Compilation as a Service for cloud-native Java mi...
DevNexus 2024: Just-In-Time Compilation as a Service for cloud-native Java mi...
 
JITServerTalk JCON World 2023.pdf
JITServerTalk JCON World 2023.pdfJITServerTalk JCON World 2023.pdf
JITServerTalk JCON World 2023.pdf
 
ClickOS_EE80777777777777777777777777777.pptx
ClickOS_EE80777777777777777777777777777.pptxClickOS_EE80777777777777777777777777777.pptx
ClickOS_EE80777777777777777777777777777.pptx
 
CloudOpen Japan - Controlling the cost of your first cloud
CloudOpen Japan - Controlling the cost of your first cloudCloudOpen Japan - Controlling the cost of your first cloud
CloudOpen Japan - Controlling the cost of your first cloud
 
JITServerTalk.pdf
JITServerTalk.pdfJITServerTalk.pdf
JITServerTalk.pdf
 
Virtualization for competitive advantage - Eric Vanderburg
Virtualization for competitive advantage - Eric VanderburgVirtualization for competitive advantage - Eric Vanderburg
Virtualization for competitive advantage - Eric Vanderburg
 
Software Defined Agility for IBM FlashSystem V9000
Software Defined Agility for IBM FlashSystem V9000Software Defined Agility for IBM FlashSystem V9000
Software Defined Agility for IBM FlashSystem V9000
 
JITServerTalk Nebraska 2023.pdf
JITServerTalk Nebraska 2023.pdfJITServerTalk Nebraska 2023.pdf
JITServerTalk Nebraska 2023.pdf
 
Cloud Computing: Safe Haven from the Data Deluge? AGBT 2011
Cloud Computing: Safe Haven from the Data Deluge? AGBT 2011Cloud Computing: Safe Haven from the Data Deluge? AGBT 2011
Cloud Computing: Safe Haven from the Data Deluge? AGBT 2011
 
Presentation citrix cloud platform for infrastructure as a service
Presentation   citrix cloud platform for infrastructure as a servicePresentation   citrix cloud platform for infrastructure as a service
Presentation citrix cloud platform for infrastructure as a service
 
Cloud stack overview
Cloud stack overviewCloud stack overview
Cloud stack overview
 
(CMP402) Amazon EC2 Instances Deep Dive
(CMP402) Amazon EC2 Instances Deep Dive(CMP402) Amazon EC2 Instances Deep Dive
(CMP402) Amazon EC2 Instances Deep Dive
 
1 virtual machines
1 virtual machines1 virtual machines
1 virtual machines
 
20191201 kubernetes managed weblogic revival - part 1
20191201 kubernetes managed weblogic revival - part 120191201 kubernetes managed weblogic revival - part 1
20191201 kubernetes managed weblogic revival - part 1
 
Geographically Distributed Multi-Master MySQL Clusters
Geographically Distributed Multi-Master MySQL ClustersGeographically Distributed Multi-Master MySQL Clusters
Geographically Distributed Multi-Master MySQL Clusters
 
The impact of cloud NSBCon NY by Yves Goeleven
The impact of cloud NSBCon NY by Yves GoelevenThe impact of cloud NSBCon NY by Yves Goeleven
The impact of cloud NSBCon NY by Yves Goeleven
 

More from Hwanju Kim

CPU Scheduling for Virtual Desktop Infrastructure
CPU Scheduling for Virtual Desktop InfrastructureCPU Scheduling for Virtual Desktop Infrastructure
CPU Scheduling for Virtual Desktop InfrastructureHwanju Kim
 
4. Memory virtualization and management
4. Memory virtualization and management4. Memory virtualization and management
4. Memory virtualization and managementHwanju Kim
 
3. CPU virtualization and scheduling
3. CPU virtualization and scheduling3. CPU virtualization and scheduling
3. CPU virtualization and schedulingHwanju Kim
 
1.Introduction to virtualization
1.Introduction to virtualization1.Introduction to virtualization
1.Introduction to virtualizationHwanju Kim
 
Demand-Based Coordinated Scheduling for SMP VMs
Demand-Based Coordinated Scheduling for SMP VMsDemand-Based Coordinated Scheduling for SMP VMs
Demand-Based Coordinated Scheduling for SMP VMsHwanju Kim
 
Scheduler Support for Video-oriented Multimedia on Client-side Virtualization
Scheduler Support for Video-oriented Multimedia on Client-side VirtualizationScheduler Support for Video-oriented Multimedia on Client-side Virtualization
Scheduler Support for Video-oriented Multimedia on Client-side VirtualizationHwanju Kim
 
Task-aware Virtual Machine Scheduling for I/O Performance
Task-aware Virtual Machine Scheduling for I/O PerformanceTask-aware Virtual Machine Scheduling for I/O Performance
Task-aware Virtual Machine Scheduling for I/O PerformanceHwanju Kim
 

More from Hwanju Kim (8)

CPU Scheduling for Virtual Desktop Infrastructure
CPU Scheduling for Virtual Desktop InfrastructureCPU Scheduling for Virtual Desktop Infrastructure
CPU Scheduling for Virtual Desktop Infrastructure
 
4. Memory virtualization and management
4. Memory virtualization and management4. Memory virtualization and management
4. Memory virtualization and management
 
3. CPU virtualization and scheduling
3. CPU virtualization and scheduling3. CPU virtualization and scheduling
3. CPU virtualization and scheduling
 
2. OS vs. VMM
2. OS vs. VMM2. OS vs. VMM
2. OS vs. VMM
 
1.Introduction to virtualization
1.Introduction to virtualization1.Introduction to virtualization
1.Introduction to virtualization
 
Demand-Based Coordinated Scheduling for SMP VMs
Demand-Based Coordinated Scheduling for SMP VMsDemand-Based Coordinated Scheduling for SMP VMs
Demand-Based Coordinated Scheduling for SMP VMs
 
Scheduler Support for Video-oriented Multimedia on Client-side Virtualization
Scheduler Support for Video-oriented Multimedia on Client-side VirtualizationScheduler Support for Video-oriented Multimedia on Client-side Virtualization
Scheduler Support for Video-oriented Multimedia on Client-side Virtualization
 
Task-aware Virtual Machine Scheduling for I/O Performance
Task-aware Virtual Machine Scheduling for I/O PerformanceTask-aware Virtual Machine Scheduling for I/O Performance
Task-aware Virtual Machine Scheduling for I/O Performance
 

Recently uploaded

IVE Industry Focused Event - Defence Sector 2024
IVE Industry Focused Event - Defence Sector 2024IVE Industry Focused Event - Defence Sector 2024
IVE Industry Focused Event - Defence Sector 2024Mark Billinghurst
 
Oxy acetylene welding presentation note.
Oxy acetylene welding presentation note.Oxy acetylene welding presentation note.
Oxy acetylene welding presentation note.eptoze12
 
GDSC ASEB Gen AI study jams presentation
GDSC ASEB Gen AI study jams presentationGDSC ASEB Gen AI study jams presentation
GDSC ASEB Gen AI study jams presentationGDSCAESB
 
Risk Assessment For Installation of Drainage Pipes.pdf
Risk Assessment For Installation of Drainage Pipes.pdfRisk Assessment For Installation of Drainage Pipes.pdf
Risk Assessment For Installation of Drainage Pipes.pdfROCENODodongVILLACER
 
UNIT III ANALOG ELECTRONICS (BASIC ELECTRONICS)
UNIT III ANALOG ELECTRONICS (BASIC ELECTRONICS)UNIT III ANALOG ELECTRONICS (BASIC ELECTRONICS)
UNIT III ANALOG ELECTRONICS (BASIC ELECTRONICS)Dr SOUNDIRARAJ N
 
CCS355 Neural Network & Deep Learning Unit II Notes with Question bank .pdf
CCS355 Neural Network & Deep Learning Unit II Notes with Question bank .pdfCCS355 Neural Network & Deep Learning Unit II Notes with Question bank .pdf
CCS355 Neural Network & Deep Learning Unit II Notes with Question bank .pdfAsst.prof M.Gokilavani
 
Electronically Controlled suspensions system .pdf
Electronically Controlled suspensions system .pdfElectronically Controlled suspensions system .pdf
Electronically Controlled suspensions system .pdfme23b1001
 
pipeline in computer architecture design
pipeline in computer architecture  designpipeline in computer architecture  design
pipeline in computer architecture designssuser87fa0c1
 
Churning of Butter, Factors affecting .
Churning of Butter, Factors affecting  .Churning of Butter, Factors affecting  .
Churning of Butter, Factors affecting .Satyam Kumar
 
Past, Present and Future of Generative AI
Past, Present and Future of Generative AIPast, Present and Future of Generative AI
Past, Present and Future of Generative AIabhishek36461
 
Work Experience-Dalton Park.pptxfvvvvvvv
Work Experience-Dalton Park.pptxfvvvvvvvWork Experience-Dalton Park.pptxfvvvvvvv
Work Experience-Dalton Park.pptxfvvvvvvvLewisJB
 
HARMONY IN THE NATURE AND EXISTENCE - Unit-IV
HARMONY IN THE NATURE AND EXISTENCE - Unit-IVHARMONY IN THE NATURE AND EXISTENCE - Unit-IV
HARMONY IN THE NATURE AND EXISTENCE - Unit-IVRajaP95
 
Concrete Mix Design - IS 10262-2019 - .pptx
Concrete Mix Design - IS 10262-2019 - .pptxConcrete Mix Design - IS 10262-2019 - .pptx
Concrete Mix Design - IS 10262-2019 - .pptxKartikeyaDwivedi3
 
Gurgaon ✡️9711147426✨Call In girls Gurgaon Sector 51 escort service
Gurgaon ✡️9711147426✨Call In girls Gurgaon Sector 51 escort serviceGurgaon ✡️9711147426✨Call In girls Gurgaon Sector 51 escort service
Gurgaon ✡️9711147426✨Call In girls Gurgaon Sector 51 escort servicejennyeacort
 
What are the advantages and disadvantages of membrane structures.pptx
What are the advantages and disadvantages of membrane structures.pptxWhat are the advantages and disadvantages of membrane structures.pptx
What are the advantages and disadvantages of membrane structures.pptxwendy cai
 
Study on Air-Water & Water-Water Heat Exchange in a Finned Tube Exchanger
Study on Air-Water & Water-Water Heat Exchange in a Finned Tube ExchangerStudy on Air-Water & Water-Water Heat Exchange in a Finned Tube Exchanger
Study on Air-Water & Water-Water Heat Exchange in a Finned Tube ExchangerAnamika Sarkar
 

Recently uploaded (20)

IVE Industry Focused Event - Defence Sector 2024
IVE Industry Focused Event - Defence Sector 2024IVE Industry Focused Event - Defence Sector 2024
IVE Industry Focused Event - Defence Sector 2024
 
Oxy acetylene welding presentation note.
Oxy acetylene welding presentation note.Oxy acetylene welding presentation note.
Oxy acetylene welding presentation note.
 
GDSC ASEB Gen AI study jams presentation
GDSC ASEB Gen AI study jams presentationGDSC ASEB Gen AI study jams presentation
GDSC ASEB Gen AI study jams presentation
 
Risk Assessment For Installation of Drainage Pipes.pdf
Risk Assessment For Installation of Drainage Pipes.pdfRisk Assessment For Installation of Drainage Pipes.pdf
Risk Assessment For Installation of Drainage Pipes.pdf
 
UNIT III ANALOG ELECTRONICS (BASIC ELECTRONICS)
UNIT III ANALOG ELECTRONICS (BASIC ELECTRONICS)UNIT III ANALOG ELECTRONICS (BASIC ELECTRONICS)
UNIT III ANALOG ELECTRONICS (BASIC ELECTRONICS)
 
CCS355 Neural Network & Deep Learning Unit II Notes with Question bank .pdf
CCS355 Neural Network & Deep Learning Unit II Notes with Question bank .pdfCCS355 Neural Network & Deep Learning Unit II Notes with Question bank .pdf
CCS355 Neural Network & Deep Learning Unit II Notes with Question bank .pdf
 
Electronically Controlled suspensions system .pdf
Electronically Controlled suspensions system .pdfElectronically Controlled suspensions system .pdf
Electronically Controlled suspensions system .pdf
 
pipeline in computer architecture design
pipeline in computer architecture  designpipeline in computer architecture  design
pipeline in computer architecture design
 
Churning of Butter, Factors affecting .
Churning of Butter, Factors affecting  .Churning of Butter, Factors affecting  .
Churning of Butter, Factors affecting .
 
Past, Present and Future of Generative AI
Past, Present and Future of Generative AIPast, Present and Future of Generative AI
Past, Present and Future of Generative AI
 
🔝9953056974🔝!!-YOUNG call girls in Rajendra Nagar Escort rvice Shot 2000 nigh...
🔝9953056974🔝!!-YOUNG call girls in Rajendra Nagar Escort rvice Shot 2000 nigh...🔝9953056974🔝!!-YOUNG call girls in Rajendra Nagar Escort rvice Shot 2000 nigh...
🔝9953056974🔝!!-YOUNG call girls in Rajendra Nagar Escort rvice Shot 2000 nigh...
 
Work Experience-Dalton Park.pptxfvvvvvvv
Work Experience-Dalton Park.pptxfvvvvvvvWork Experience-Dalton Park.pptxfvvvvvvv
Work Experience-Dalton Park.pptxfvvvvvvv
 
Call Us -/9953056974- Call Girls In Vikaspuri-/- Delhi NCR
Call Us -/9953056974- Call Girls In Vikaspuri-/- Delhi NCRCall Us -/9953056974- Call Girls In Vikaspuri-/- Delhi NCR
Call Us -/9953056974- Call Girls In Vikaspuri-/- Delhi NCR
 
HARMONY IN THE NATURE AND EXISTENCE - Unit-IV
HARMONY IN THE NATURE AND EXISTENCE - Unit-IVHARMONY IN THE NATURE AND EXISTENCE - Unit-IV
HARMONY IN THE NATURE AND EXISTENCE - Unit-IV
 
Concrete Mix Design - IS 10262-2019 - .pptx
Concrete Mix Design - IS 10262-2019 - .pptxConcrete Mix Design - IS 10262-2019 - .pptx
Concrete Mix Design - IS 10262-2019 - .pptx
 
Design and analysis of solar grass cutter.pdf
Design and analysis of solar grass cutter.pdfDesign and analysis of solar grass cutter.pdf
Design and analysis of solar grass cutter.pdf
 
POWER SYSTEMS-1 Complete notes examples
POWER SYSTEMS-1 Complete notes  examplesPOWER SYSTEMS-1 Complete notes  examples
POWER SYSTEMS-1 Complete notes examples
 
Gurgaon ✡️9711147426✨Call In girls Gurgaon Sector 51 escort service
Gurgaon ✡️9711147426✨Call In girls Gurgaon Sector 51 escort serviceGurgaon ✡️9711147426✨Call In girls Gurgaon Sector 51 escort service
Gurgaon ✡️9711147426✨Call In girls Gurgaon Sector 51 escort service
 
What are the advantages and disadvantages of membrane structures.pptx
What are the advantages and disadvantages of membrane structures.pptxWhat are the advantages and disadvantages of membrane structures.pptx
What are the advantages and disadvantages of membrane structures.pptx
 
Study on Air-Water & Water-Water Heat Exchange in a Finned Tube Exchanger
Study on Air-Water & Water-Water Heat Exchange in a Finned Tube ExchangerStudy on Air-Water & Water-Water Heat Exchange in a Finned Tube Exchanger
Study on Air-Water & Water-Water Heat Exchange in a Finned Tube Exchanger
 

6. Live VM migration

  • 2. Outline • Live VM migration • Use cases • Live migration mechanisms • Pre-copy live migration • Post-copy live migration • Related research • Energy savings of idle desktops using virtualization • LiteGreen • Jettison • Cloud Micro-Elasticity via VM State Coloring • Kaleidoscope 2/38
  • 4. Live VM Migration • Live VM relocation • Lively synchronizing memory contents including CPU states • Storage is shared in LAN (e.g., NAS) Network Storage Configuration Data VM Memory ContentMemory Sync VM User 4/38
  • 5. What is “Live”? • Migration metrics • Total migration time • Time elapsed until all VM states including CPU and memory are transferred • Load is changed (balanced) after this time • Downtime • Time elapsed while a VM is being stopped • Service is unavailable during downtime • What is live migration? • Migration with near-zero downtime 5/38
  • 6. How to Migrate a VM • How to synchronize memory contents? • Stop-and-copy • Stop the source VM • Copy its memory contents over network • Start the destination VM • Pre-copy • Copy memory contents over network • Keep copying only dirty pages iteratively • Stop the source VM if # of dirty pages is under threshold • Copy remaining dirty pages • Start the destination VM • Post-copy • Stop the source VM • Copy CPU states and page tables over network • Start the destination VM • Copy its memory contents on demand Downtime ∝ Memory size Not live! Near-zero downtime Live! 6/38
  • 7. Pre-copy vs. Post-copy • Pros and cons Pre-copy migration Post-copy migration Eager copy of source VM’s memory Lazy copy of source VM’s memory - Longer and unpredictable downtime depending on writable working set + Shorter downtime + Shorter total migration time - Longer total migration time + High performance after migration - Low performance after migration due to network page fault - Waste network bandwidth by pages that will not touched by a destination VM + Effective use of network bandwidth 7/38
  • 8. Pre-copy vs. Post-copy • Trade-off Totalmigrationtime Downtime Post-copy Stop-and-copy Pre-copy Live Since pre-copy live migration is good for downtime and migration time, it has been used in most VMMs • Overhead after migration can be effectively reduced by prefetching • Suitable for VM forking and microsleep 8/38
  • 9. Pre-copy Live Migration • “Live migration of Virtual Machines [NSDI’05]” 9/38
  • 10. Post-copy Live Migration • “Post-Copy Live Migration of Virtual Machines [VEE’09]” • Main issue: How to reduce runtime overheads after post-copy migration Prepaging (prefetching) policy Bubbling with single pivot Bubbling with multiple pivots 10/38
  • 11. ENERGY SAVINGS OF IDLE DESKTOPS USING VIRTUALIZATION Related research 11
  • 12. Introduction • How serious is desktop energy consumption? Source: Greener PCs for the enterprises 12/38
  • 13. Introduction • Why nontrivial for desktop energy savings? VS Users don’t want ongoing jobs to be disrupted even when away Great savings when away Roughly 60% of office desktop PCs are left on continuously 13/38
  • 14. Naïve Method • Sleep • ACPI S3 and S4 states • S3 – standby (suspend to RAM) • S4 – hibernate (suspend to disk) • Pros. • Significant energy savings • Cons. • Losing network presence I expect the torrent download to complete after drinking! So, don’t sleep!!!! How to save energy with handling user’s ongoing or potential tasks 14/38
  • 15. Existing Methods • Proxy-based Approach • WoL (Wake-On-Lan) proxy • Same subnet, known MAC addresses, manual operation • Protocol proxy [NSDI‘09, USENIX’10] • Triggered by a filtered subset of the incoming traffic • Listening network ports, user input • Explicit specification before sleep • Application proxy [NSDI‘09] • Application-specific stubs • Complexity for creating each application stub 15/38
  • 16. LiteGreen Project (Mircosoft) • LiteGreen: Saving Energy in Networked Desktops Using Virtualization [USENIX’10] • Achieving the conflicting goals • Energy saving and continuous computing • Eliminating complexity from protocol- or application- specific approaches • locating a desktop in local desktop • for good user experiences • consolidating idle desktops in a server • for energy savings VM Live migration!! 16/38
  • 18. How LiteGreen Works • Operations Hypervisor stub VM Hypervisor controller VM VM VM LiteGreen Server Desktop RDP Client Live migration makes a desktop “always on” 18/38
  • 20. Problems of Full VM Migration • Excessive network bandwidth for migration • VM memory size + alpha (dirty block copies) • e.g., about 4.27GB for 4GB VM • “Boot storm” (after lunch) • Long migration time • Delayed sleep • e.g., 38sec for 1VM, 253sec for 8 VMs • Less energy savings • Full VM migration after ballooning  ballooning requires considerable time and I/O • Consolidation aborted by short idle time • Long resume time • Poor user experience 20/38
  • 21. Jettison • Jettison: Efficient Idle Desktop Consolidation with Partial VM Migration [EuroSys’12] • Goals • Quick resume • Good user experience • Conservation of the network resources • Efficiency and scalability • Cost effective • Reduction in TCO by energy savings • Idea • “Partial VM migration” with fetching required parts on demand 21/38
  • 22. Partial VM Migration • Jettison Hypervisor stub Hypervisor controller VM VM VM Jettison Server Desktop VM Sleep (S3) Wake-on-LAN VM 1. Idleness detection 2. Consolidation 4. On-demand fetch 3. Microsleep 5. Reintegration Procedure 22/38
  • 23. State Prefetch • Prefetch for increasing inter-arrivals of remote faults • Hoarding • Based on fetched frame sequence of a previous migration • On-demand prefetch • Based on spatial locality 23/38
  • 24. State Prefetch • Trace-driven offline analysis • Page access traces from a user VM consolidated 58 times On-demand prefetch works well with 20 page window 24/38
  • 25. Budget Analysis • Full vs. Parital VM migration • Assuming 16GiB memory SunFire X2250 • USD 6099 • Full VM migration • 33.95 USD / desktop / year • 33.95 x 4 VMs x 3 years = USD 407.40 • Partial VM migration • 37.35 USD / desktop / year • 37.35 x 98 VMs x 3 years = USD 10,980.90 25/38
  • 26. KALEIDOSCOPE: CLOUD MICRO- ELASTICITY VIA VM STATE COLORING Related Research 26
  • 27. Elasticity of Clouds • Ideal elasticity: Pay-per-use model • Achieves both QoS and efficient resource utilization Source: http://astadiaemea.wordpress.com/2010/06/ 27/38
  • 28. What Matters for Elasticity? • Granularity • A unit of service delivery and billing • VM as a unit • IaaS (e.g., Amazon EC2) • Coarse granularity • A VM booting from scratch • QoS • Well-known trade-off against resource utilization • Conservative elasticity • High QoS, but inefficient resource utilization • Aggressive elasticity • Low QoS, but efficient resource utilization Ideal Cloud! How about QoS? Too slow for ideal elasticity 28/38
  • 29. QoS in Clouds • Dynamic adjustment of worker VM pool • Amazon EC2 • Auto Scaling • Elastic Load Balancing • Load balancing using elasticity • Load > TH • Inflate VM pool by requesting additional VMs • Load < TL • Deflate VM pool by returning unnecessary VMs • High threshold • Achieves aggressive elasticity for efficient resource utilization • Requires fast VM instantiation for QoS 29/38
  • 30. Elasticity Needs • AT&T’s hosting in January 2010 Needs for elasticity Short-lived workers 30/38
  • 31. Problems of Current Clouds • Slow VM instantiation • Average 2min to boot a VM (Amazon EC2) • Very fluctuating latencies • Cold status of new VMs • Initially empty OS caches • Performance degradation during peak load • Inefficient resource utilization of new VMs • Full memory allocation during short-lived VMs that require smaller working set 31/38
  • 32. Micro-Elasticity • Goals • Fast VM instantiation • VM cloning: SnowFlock [Eurosys’09] • Efficient memory utilization for short-lived VM • On-demand resource allocation • Warm status of new VMs • Prefetching related data: VM state coloring Color-based fractional VM cloning 32/38
  • 33. Live VM Cloning • Trade-off between cloning techniques Post-copy cloning Pre-copy cloning SnowFlock [EuroSys’09] Like live migration Lazy copy of parent’s memory Eager copy of parent’s memory Short cloning time Long and unpredictable cloning time Low performance after cloning due to the cold status High performance after cloning due to the warm status Effective use of network bandwidth & Possibility of memory savings Waste of memory and network bandwidth by pages that will not touched by clone VMs 33/38
  • 34. VM State Coloring • Effective VM memory prefetching scheme • Assuming that locality exist within a related region • Partitioning VM memory into semantically related regions • Methods • Architecture-based coloring • Introspective coloring VM memory = Uniform binary state VM state coloring 34/38
  • 35. VM State Coloring • Color map example • SPECweb Support workload • Interspersing of different colors in the physical memory space of the VM Yello –page cache Light blue – user data Dark blue – kernel data Light red – user code Dark red – kernel code Black - free 35/38
  • 36. VM State Coloring • Benefits of per-color prefetching against color- blind prefetching • Accuracy • Fewer wasted fetches of unneeded pages • Efficiency • Less page faults • Per-color prefetch tuning 36/38
  • 37. Implications for Clouds • QoS and resource use • Kaleidoscope with TH=90% outperforms Elastic Clouds with TH=50% 37/38
  • 38. Summary • Live migration is a key technique of virtualization • Pre-copy live migration • Working well for general workloads • No performance degradation after migration • Used by most VMMs • Post-copy live migration • On-demand migration • Efficient bandwidth usage • Strong for write-intensive workloads • Assisted by prefetching 38/38