Summary of "Lec 19: VM Placement and VM Migration"

Overview

Topic: VM placement (choosing hosts for virtual machines) and VM migration (moving VMs between hosts).
Purpose: maximize resource utilization and performance, minimize energy use and costs, and reduce SLA violations, latency, congestion, and failures.

VM placement

VM placement: selecting the most suitable physical host from a pool of servers based on VM requirements (CPU, RAM, network).

Goals:

Energy efficiency
Load balancing
Minimize SLA and network issues
Cost reduction

Placement schemes:

Static placement
- VM-to-host mapping is fixed for the VM’s lifetime (no host changes).
Dynamic placement
- VM host can change over time based on conditions.
- Reactive: migrate only after problems appear (e.g., overload, maintenance, power issues).
- Proactive: migrate before problems occur based on predictions or thresholds.

VM migration

VM migration: moving a VM (or its storage) from one physical host, storage location, or data center to another to optimize resources, enable maintenance, or balance load.

Two main migration types:

Hot (live) migration

VM keeps running during most of the process; the goal is minimal downtime (usually milliseconds to seconds).
Typical requirements:
- Shared storage between source and target hosts.
- CPU compatibility checks between hosts.
Overhead:
- Iterative memory copying can cause repeated copying of pages that get dirtied during transfer.
Typical process stages (often numbered 0–5):
1. Pre-migration: select target host, mirror block devices, ensure resources are available.
2. Reservation: initialize container on target and reserve CPU/RAM/storage.
3. Iterative pre-copy: iteratively copy memory pages from source to target (shadow paging; repeated rounds for pages dirtied during copy).
4. Stop-and-copy: briefly suspend the VM on the source, copy remaining CPU/memory state, and update ARP/network to route to target (small downtime).
5. Commitment: release resources on the source host and delete the old instance.
6. Activation: start/resume the VM on the target and reconnect devices/network; users continue.

Cold (powered-off) migration

The VM is stopped; its entire state (disk, RAM image, config, logs) is moved, then the VM is started on the target.
Does not require shared storage or the same CPU compatibility checks required by live migration.
Simpler process but incurs full downtime for the duration of the move.
Useful for moving associated disks between datastores and across data centers.

Key comparisons

Live (hot) migration
- Pros: lower downtime.
- Cons: needs shared storage and CPU compatibility checks; more complex due to iterative copying and coordination.
Cold migration
- Pros: simpler; can move disks and cross data centers without shared storage.
- Cons: higher downtime.

Type of content

This is a tutorial/lecture covering definitions, goals, placement schemes, migration methods, detailed live-migration steps, and practical requirements/overheads.