Virtual Machine Live Migration

• Virtual Machine (VM) live migration involves moving VM instances between physical hosts with minimal downtime, without disrupting services.
• Services remain unaware of the migration process, maintaining network connections of the guest OS.
• The VM is treated as a black box during migration.

Why VM Migration?

• Online migration can vacate a VM Host system, target a specific VM Host, balance host workloads, and optimize physical resource utilization.
• Offline migration uses fewer resources than online migration on the source and target VM Hosts.
• Offline migration is necessary if there are different processor types preventing online migration, if the source VM may not support online migration, or if the VM has stopped and configuration information needs to be moved offline.

Examples of VM Migration

• Amazon-AWS provides VM migration from one EC2 to another, or from EC2 to Virtual Private Cloud (VPC).
• VMWare-vMotion allows moving a VM from one physical server to another.
• RackSpace-Migration services facilitates migration from cloud service provider to in-house server, or from one cloud service provider to another.
• Microsoft Hyper-V virtualizes Windows Server and can migrate a VM running on "Virtual Server 2005 R2 SP1" to Hyper-V.
• OpenStack supports both live and non-live migrations for cloud administrators.
• Red Hat enables live migration of VMs for maintenance, load balancing, or power saving.

OS-Level Migration

• OS-level migration avoids 'residual dependencies.'
• The original host can be powered off/put to sleep once migration is complete.
• In-memory state can be transferred consistently and efficiently, eliminating reconnection needs for applications like media streaming.
• OS-level migration allows a separation of concerns between users and operators, where users control software/services within their VM, while operators remain agnostic.

Design Challenges

• Minimize service downtime.
• Minimize migration duration.
• Avoid disrupting the running service.

Virtual Machine Live Migration

• Purposes:
  • Eliminate hot-spots to free up resources.
  • Defragment a data center or obtain resources.
• Techniques:
  • Pre-copy live migration.
  • Post-copy live migration.
  • Hybrid live migration.

Pre-copy Migration

• Step 1: Establish layer-2 connectivity between data centers if needed.
• Step 2: Transfer the application's disk state if storage is not shared.
• Step 3: Transfer the memory state of the application to a server in Data Center B while it continues running without interruption.
• Step 4: Pause the application briefly for the final memory and processor state transition to Data Center B once the disk and memory states have been transferred.

Disk State Migration

• The initial copy is made asynchronously.
• Synchronous replication occurs once the remote disk is stable.
• Disk updates are continuously propagated to the remote disk.

Transferring Memory State

• A "Pre-copy" mechanism is used to iteratively copy the memory contents.
• The VM is paused at the end to copy the final memory state.
• The time required for memory transfer depends on RAM allocation, working set size, write rate, and available bandwidth.

Traditional vs. Smart Memory Transfer

• The default Xen migration algorithm iterates until only a few pages remain, more than 3 times the VM's memory is sent, or a limit of 30 iterations is reached, then the VM is paused and all remaining pages are sent.
• Smart Stop and Copy mitigates issues of long iterations by limiting transferred data.
• Migration algorithms can intelligently determine when to stop iterating.
• A heuristic that tracks the number of pages remaining to be sent.
• Looks for local minima in the current history window
• Greedy approach works best for window size 5

Page Deltas

• After the first iteration, most pages are transferred.
• Reduces bandwidth consumed during migration by sending only the changes that happened in the pages.

Design Example

• Stage 0: Pre-Migration: Active VM on Host A, alternate host pre-selected, block devices mirrored, resources maintained.
• Stage 1: Reservation: Initialize a container on the target host.
• Stage 2: Iterative Pre-copy: Shadow paging enabled, dirty pages copied in rounds.
• Stage 3: Stop and copy: VM suspended on Host A, ARP generated to redirect traffic, remaining state synchronized to Host B.
• Stage 4: Commitment: VM state is released on Host A.
• Stage 5: Activation: VM starts on Host B, connects to local devices, and resumes operation.

Post-copy Migration

• Involves stopping the VM, moving non-memory states to the destination, and beginning execution on the new machine.
• During page faults, the required page is copied from the source machine, with background copying ensuring all pages are copied.
• Is useful when avoiding latency.

Disadvantages of Pre-Copy and Post-Copy

• Pre-copy issues include long migration times, large data transfer, and degrading application performance.
• Post-copy can cause network-bound page faults, potential failures, and performance degradation.

Challenges and Impacts of VM Migration

• Minimizing service downtime and migration duration, and avoiding disruption to running services.
• Energy consumption is a consideration.

Traffic Contention

• Migration traffic competes for bandwidth, especially when transmitting in the same direction.

Traffic Contention Examples

• Outgoing traffic contends with pre-copy migration in source hosts.
• Incoming traffic contends with post-copy migration in destination hosts.

Traffic Contention Solutions

• Setting a limit on total migration time, data amount, or transfer rate.
• Trade-off: reducing overhead of pre-copy may increase downtime, while offering more bandwidth to the VM may increase total migration time.

Traffic-Sensitive Migration

• Select migration options to minimize affect on application.
• Fully utilize available network bandwidth.
• Avoid traffic contention, increases throughput.

Basic Concept

• Traffic Profile requires recording the rate of network traffic which is used to calculate traffic contention.
• A combination of pre-copy and post-copy is often used.

Prospect

• Extending traffic-sensitive migration by moving VMs to differing hosts.
• Calculate traffic contention, in destination hosts.
• Select destination host and minimizing network contention.