AWS Virtual IP Cross-Availability Zone Architecture Overview
Please refer to the diagram shown to the right for the remaining discussion.
This diagram depicts a typical HA deployment but is not the only possible design. In fact, SoftNAS SNAP HA™ can be deployed with dual controllers located within a single AZ on a VPC (there is no requirement to split controllers across AZs, but it is a recommended best practice for maximum availability).
We see a VPC created in a /16 network in AWS US East (Virginia) data center, with subnets allocated in Zone A and Zone B. This topology provides the best overall redundancy and availability within the AWS AZ architecture.
Two SoftNAS® controller EC2 instances are deployed - one per AZ. If optional private subnets are configured in one or more AZs, they will also have access to the Virtual IP(VIP) for NFS client storage access via NFS, CIFS and iSCSI protocols.
The drawing shows SNAP HA™ replication traffic flowing from Controller A to Controller B. This traffic is allocated to interface 0. Interface 0 is also used for administration using the SoftNAS StorageCenter™ Administrative Interface. Block replication keeps a warm copy of the data from node A on node B, in case a failover is necessary.
The drawing shows two orange arrows emanating from an orange and white circle, which represents the Virtual IP. The black lock symbol represents the EC2 Security Group associated with the Virtual IP. The shadowed orange arrow represents re-routed storage requests flowing to Controller B after an automatic failover or manual takeover. This Virtual IP must be in a completely separate CIDR block from the two instance IPs.
When an automatic failover or manual takeover occurs, NAS traffic is re-routed via the Virtual IP from Controller A to Controller B, as indicated by in the diagram above. When a Virtual IP switches over from one controller to another, NAS client traffic is rapidly re-routed to the new controller, typically in just a few seconds. NAS clients typically experience a brief switch-over delay of up to 20 seconds or so, and automatically reconnect after the switch-over event takes place.