It's very difficult to develop a system with zero downtime. As a GIS is typically multiple tiers, (web tier, application tier, data tier, etc), all of them have to be HA and fully resilient in order to get zero downtime. On top of that, your IT staff needs to be "HA" as well. If you experience an issue, your staff needs to be immediately available to address the problem. Often times a separate a data center is configured in the event that your original data center has some type of catastrophic failure like a natural disaster. You'd use machine replication or the DR tool to move data and content over to your standby. All URLs within your deployment, admin URL used for federating, privatePortalURL for Server to Portal communication, front-end URLs, need to be through a fault-tolerant load balancer. You need to have monitoring in place to immediately detect and resolve issues.
Ultimately, there's a lot to consider when developing a system with high uptime. The software does its best to be highly available, (multi-machine Server sites, Portal and Data Store), but minimizing downtime it goes far beyond software and isn't a short conversation to be had over a forum. It's best to seek some type of professional services engagement who can evaluate your current infrastucture and develop an HA plan that fits your needs.
