What is load balancing?

Two benefits arrive together. Capacity beyond what one machine can handle, and survival when one of them dies.

The distribution rule can be simple or clever. Round the list in turn, send to the least busy, or keep a given user on the same machine.

That last option matters for anything holding state mid-conversation, which is why voice systems care about it more than websites do.

The balancer itself becomes the thing that must not fail, which is why serious setups run more than one of those too.

Say a voice service runs three servers behind a load balancer, and each can hold 200 live calls. On an ordinary afternoon 150 calls are in progress, about 50 per server. The balancer checks each machine every five seconds with a tiny health request. At 4:58 pm server B stops answering. After two missed checks, the balancer marks it unhealthy and sends new calls only to A and C.

Here's what people miss. Those 50 calls already on B are gone, because a live conversation can't hop from one machine to another in the middle of a sentence. Your caller on B hears silence and redials, while new callers notice nothing. Then a thunderstorm knocks out power across town and 450 people call at once. Two servers hold 400, so everyone past that gets a busy signal or a queue, depending on the design. Balancing spread the load perfectly, and capacity still ran out.

If you've set up office phones, you've met a relative of this idea. A hunt group rings extension 101, then 102, then 103. With a ring group, all of them ring at once. Both spread calls across people. Load balancing spreads work across machines, and your caller never knows it happened. Sales conversations blur the two, so when a vendor tells you "we balance your calls", stop and ask which one they mean.

Each kind has its own questions. For the people side, you care about ring order and how long a call waits before it moves on. On the machine side, ask what your caller hears if the server handling them fails. Ask whether your account has a limit on simultaneous calls, and what the number is.

At four calls an hour, none of this will ever touch you. It matters when your calls arrive in bursts, such as after a radio ad or the morning after a holiday. Bring the size of your worst burst to the conversation.

GreetKeeper answers calls that arrive together at the same time, each in its own conversation. We don't publish a maximum, so if your peaks are unusual, ask us with your own numbers.

How our platform spreads that work across servers is our problem rather than yours. What you should check on your side is the channel limit on your own phone line, because callers beyond it never reach us.

Bring your real busy-hour pattern to a demo and we will show you what the product does rather than describe it.