Menu

Load Balancers

One address in front of many servers.

A load balancer is the thing your domain points at instead of a single server. It hands each request to one of the machines behind it, checks that they are still answering, and quietly stops sending traffic to any that are not.

Health checks
a sick server stops receiving traffic
One address
your DNS points here, not at a server
SSL
terminate certificates at the balancer
No downtime
deploy one server at a time

What it is

While you have one server, that server is also your single point of failure and your ceiling. When it is busy, everybody waits; when it reboots, the site is down; and when you deploy, somebody sees the deploy.

A load balancer removes all three at once. Two or more servers sit behind one address, work is shared between them, and any one of them can be taken out for a deploy, a reboot or a bad afternoon without the site noticing.

What it does for you

Six things Load Balancers changes about how your infrastructure behaves.

Traffic spread across servers

Each request goes to one of the machines behind the balancer, so a busy hour is shared rather than landing on one box.

A failed server leaves quietly

The balancer checks each server on a schedule. One that stops answering is taken out of rotation until it recovers — without a phone call at two in the morning.

Deploy without an error page

Take one server out, update it, put it back, repeat. Visitors are served by the others throughout.

SSL handled in one place

Terminate certificates at the balancer instead of installing and renewing them on every server behind it.

Sticky sessions when you need them

Applications that keep a session in memory can pin a visitor to the server that has it, until you have moved sessions somewhere shared.

Room to grow sideways

When one server is not enough, the answer stops being "a bigger server" and becomes "another server" — which has no ceiling.

What people use it for

A site that outgrew one server
Traffic that spikes at predictable times — a sale, a results day, a campaign — served by more machines during the spike.
Zero-downtime deployment
Roll a release through the pool one server at a time, with the option to stop halfway if it goes wrong.
Redundancy for an application that matters
Two servers in the pool means a failed machine is an incident report, not an outage.

What it costs

A load balancer is priced per balancer, separately from the servers behind it. Tell us how many servers you want in the pool and we will quote it.

We would rather quote you than publish a number we have not checked. Everything on the pricing page is a figure the order form actually charges — this will join it the day the same is true here.

Questions

How many servers do I need behind it?

Two is the number that changes anything — one server behind a balancer is still one server. Beyond two, add machines as traffic asks for them.

Does my application need changing?

Usually a little. Anything the application keeps on local disk — uploaded files, sessions, caches — has to move somewhere shared, or each visitor will get a different answer depending on which server they land on. Object storage handles the files; a shared cache handles sessions.

Where does my SSL certificate go?

On the balancer, which is simpler — one certificate to install and renew rather than one per server.

Can it balance across providers?

No. The pool lives on one platform, in one region. Spreading a single application across two providers is a bigger piece of engineering than a load balancer, and usually not the one worth doing first.

Something not answered here? Ask us — it is a short answer or it is a phone call, and either is fine.

Not sure whether you need this?

Tell us what you are running and what is going wrong with it. We will say which of these actually helps — including when the answer is none of them.