Dual-stack networking runs IPv4 and IPv6 alongside each other. A dual-stack host typically has addresses from both families and selects an appropriate destination address when establishing a connection. Each connection normally uses one address family; packets are not automatically translated between IPv4 and IPv6. An IPv4-only endpoint cannot communicate directly with an IPv6-only endpoint without an intermediary such as a translating gateway or application proxy.
In the cloud, implementation starts with address planning and capability checks. Engineers allocate IPv4 and IPv6 address ranges to supported virtual networks and subnets, configure workload interfaces, and establish routes for each family. IPv4 and IPv6 routes are separate: an IPv4 default route, 0.0.0.0/0, does not provide an IPv6 default route, ::/0. Public, private, regional, and service-specific connectivity options must be checked independently because enabling IPv6 on a network does not make every attached managed service or endpoint IPv6-capable.
DNS commonly exposes both paths through A records for IPv4 and AAAA records for IPv6. Clients choose among the returned addresses according to their operating system and application behavior. Many clients use Happy Eyeballs algorithms to reduce delays when one family is slow or unavailable, but not all applications implement this behavior. Publishing an AAAA record before validating IPv6 connectivity can therefore cause intermittent failures or long connection delays even while IPv4 remains healthy.
A practical migration begins by inventorying clients, dependencies, address literals, firewall rules, proxies, monitoring tools, and hybrid connections. Introduce IPv6 into a nonproduction environment, verify application listeners and outbound dependencies, then enable selected production workloads. A web application might first expose a dual-stack load balancer while retaining IPv4-only backends, provided the load balancer supports that combination. This makes the frontend dual-stack, not the entire application path. Native dual-stack connectivity can subsequently extend to backend services where useful.
Security policy must cover both families deliberately. An IPv4 rule does not implicitly authorize or restrict equivalent IPv6 traffic, and IPv6 can create an unintended path around IPv4-only inspection or allowlists. Globally routable IPv6 addresses do not by themselves make a resource publicly reachable; routing and firewall policy still control access. Conversely, the absence of NAT is not a security defect. Use stateful filtering and, where supported, egress-only IPv6 internet connectivity when workloads require outbound access without unsolicited inbound connections. Permit the ICMPv6 messages required for correct operation, including Packet Too Big messages used by Path MTU Discovery.
Hybrid migration also requires confirming that routers, VPNs, private interconnects, routing protocols, and inspection appliances support the required IPv6 paths. Validate return routing and test realistic packet sizes rather than relying only on small probes. Logs, metrics, access controls, and application data models must handle IPv6 addresses without truncation or assumptions about IPv4 formatting.
Dual stack reduces migration risk by retaining IPv4 compatibility, but increases operational scope because both paths need testing, policy management, troubleshooting, and capacity planning. It does not immediately eliminate IPv4 address consumption or related costs. Measure success rates and latency separately by address family, and plan rollback through traffic controls or DNS changes while accounting for cached records and existing connections. DNS64 and NAT64 are translation mechanisms commonly used for IPv6-only clients reaching IPv4 destinations; they are not prerequisites for ordinary dual-stack communication.