IT 3300 : Virtualization

Storage & High Availability

Why storage matters here

• VMs and containers are only as mobile as their storage
• To move or restart a workload on another machine, the data must be
  reachable from that machine
• So virtualization and storage are joined at the hip

Ways to store data

• Directly attached (DAS) — local disk; fast, but tied to one machine
• NAS — a networked file server (NFS/SMB) on your normal network
• SAN — block storage over a dedicated storage network

NAS

• A dedicated machine that serves files over the network
• Simple and common (we'll use TrueNAS + NFS)
• Trade-offs: network traffic to one box; if it's down, storage is down

SAN

• Presents raw block devices, not files
• Uses its own network — Fibre Channel or iSCSI
• Easy to grow; built for high availability and redundancy

Failover

• Automatically restart a VM elsewhere when its host crashes
• For mission-critical, "losing money when it's down" workloads
• Also called clustering or high availability
• Goal: no single point of failure (SPOF)

How failover works

• Keep the VM's state reachable (shared storage)
• On failure, another host resurrects the VM:
  • hardware definition, storage, IP/network, settings
• The hypervisor cluster monitors health and triggers the move

High availability

• Failover handles a dead VM — but what if the hardware dies?
• Use multiple hosts; hypervisors coordinate
• The VM's state must be able to move between hosts
  • which is why shared/distributed storage matters

Load balancing

• Run multiple instances of a service on separate hardware
• Split the workload across them
  • protects against SPOF
  • better use of resources
• Requires virtualized/shared storage to be practical

Server pooling

• Group hosts into a pool that provides failover and load balancing
  automatically
• The cluster decides where workloads run and moves them as needed

Modern echo

Everything here reappears later, one layer up:

• Proxmox HA = failover for VMs
• Kubernetes = failover, load balancing, and pooling for containers

Same ideas, higher in the stack.

Discussion

How does virtualization help with disaster recovery?