Floating IP and load balancer: when you need them
A floating IP and a load balancer come in when one server is no longer enough, either for reliability or for capacity. Both tools only work with two or more servers. Below: how they differ, what downtime they remove, what you need to change in your application and what they cost at hosts in the accsly catalog.
The short answer. A floating IP is an address that can be moved from one server to another in seconds. It lets traffic go to a standby server when the main one fails, without changing DNS. A load balancer spreads requests across several servers and stops sending them to a broken one. You need it when one server does not have enough capacity or when servers are updated one at a time. A site on a single VPS needs neither: backups and monitoring matter more for it.
What downtime they remove
Hosts most often promise 99.9% availability. That allows up to 43 minutes of downtime a month, and that downtime is the time it takes the host to repair or move your server. A floating IP and a load balancer cut it down to seconds, but only if there is a second server ready to take over. What a host actually guarantees and how availability is calculated is covered in the article hosting SLA and uptime.
Work out what an hour of downtime costs you. If an online store loses more in an hour than a second server costs per month, redundancy pays for itself. If the site is a simple business-card site, a backup and an outage alert are enough: see server monitoring.
Floating IP
An ordinary address is tied to one server. A floating address (hosts call it floating, reserved, failover or elastic IP) is issued separately and assigned to a server through the control panel or API. You can reassign it without powering the server off, and within seconds traffic goes to the other machine. DNS does not change, so ISPs' caches do not get in the way.
Limitations worth knowing about:
- It works only within one host, and often only within one data centre or zone. At Hetzner, for example, according to the documentation, the server must be in the same network zone as the floating IP. Such an address will not save you from an outage of the whole data centre.
- The address has to be configured in the server's system. The host routes traffic to the machine, but the server itself must accept it: the address is added to the network settings on both servers.
- Failover has to be automatic. This is usually done by keepalived: the two servers check each other over the VRRP protocol, and when the main one fails, the standby calls the host's API and takes over the address. In cloud networks keepalived is set up in unicast mode, where the servers contact each other directly by address.
A floating IP usually costs the same as an extra IPv4 address. Here is how it is recorded in the accsly catalog on 14 November 2026, from the hosts' own pages:
| Host | What it is called | Price per month |
|---|---|---|
| REG.RU | public floating IP | ₽187.38 (₽0.26 per hour) |
| Fotbo | Floating IP | €1.50 excl. VAT |
| Cherry Servers | Floating IP | from €2 |
| RareCloud | Reserved IP | €2 |
| TransIP | floating IPv4 in the OpenStack cloud | €2.77 |
| efect.ro | Floating IP | €3 |
| netuy.net | floating IPv4 | $1.31 |
| Lunanode | floating IP | $1 |
| Host Techno | failover IP | $3 |
For comparison: an ordinary extra IPv4 address at hosts in the catalog costs ₽150, €2 or $3 a month at the middle of the market, depending on the host's currency. Why IPv4 addresses cost money is explained in the article IPv4 and IPv6 on a server.
Load balancer
A load balancer accepts all requests and hands them out to several application servers. It regularly checks whether the servers are alive and takes those that do not respond out of rotation. This gives you three things: more capacity than one server has, survival when one of the servers fails, and updates without downtime, since servers are taken out of rotation and updated one at a time.
Load balancers come in two kinds. A managed load balancer from the host is a service with its own address and settings in the control panel: in the accsly catalog its price is recorded, for example, at Fotbo (€7.80 a month excl. VAT) and MojoHost ($10 a month). Your own load balancer is a separate VPS with nginx or HAProxy. It is cheaper and more flexible, but it becomes a single point of failure itself. That is why two load balancers are set up, with a floating IP switching between them: the classic setup that combines both tools from this article.
If you build your own, keep in mind how the programs differ. Free nginx learns that a server is down only from failed visitor requests: according to the documentation, after one error the server is taken out of rotation for 10 seconds, and periodic health checks are available only in the paid version. HAProxy can check servers regularly on its own, before a visitor ever reaches them.
What your application must support
A load balancer is useless if the application is built for a single server. Before adding a second server, check the following:
- Sessions are stored in Redis or the database, not in files on disk. Otherwise users will be logged out whenever a request lands on another server.
- Uploaded files live in shared storage, such as object storage, not on one server's disk.
- The database is shared by all application servers, or replication is set up. How databases work on a VPS is covered in the article PostgreSQL and MySQL on a VPS.
- Scheduled jobs run on one server only; otherwise emails and reports will be duplicated.
- The visitor's real address is taken from the header passed on by the load balancer, not from the connection address. Otherwise logs, IP-based limits and brute-force protection will only see the load balancer's address.
What about just changing DNS?
Several A records for one domain (DNS round robin) spread visitors across servers but do not remove a failed one: some visitors will keep hitting the dead address. Failover through DNS works, but slowly: even with a TTL of 60 seconds, some software keeps the old address longer. On the other hand, DNS is the only one of the three methods that can switch traffic between different hosts and data centres. How to move between hosts via DNS without downtime is covered in the article moving a VPS to another host.
What to choose
| Situation | What you need |
|---|---|
| One site, a few dozen minutes of downtime is tolerable | None of this. Backups, monitoring, a snapshot before updates |
| Downtime is expensive, but one server handles the load | A standby server and a floating IP with keepalived |
| One server does not have enough capacity | A load balancer and two or more application servers |
| You need both capacity and reliability | The host's managed load balancer, or two of your own with a floating IP |
| You need to survive the failure of a whole data centre | Servers in different data centres and failover through DNS |
Floating IPs and managed load balancers are offered mostly by cloud hosts that let you manage resources through an API; see the selections VPS with API management and VPS with hourly billing. How a cloud server differs from an ordinary VPS is covered in the article cloud server or VPS.
What to ask your host
- Is there a floating IP, and within what limits can it be reassigned: within one data centre or between them?
- Can the address be reassigned through the API, and how long does it take?
- Is there a managed load balancer, how much does it cost, does it run health checks on the servers and does it accept TLS certificates?
- Is there a free private network between servers, so the load balancer reaches them without going over the internet?