Nova Wholesale Bridge“Carrier-grade” and “cloud” get used as though they mean the same thing. They do not. A service can be entirely cloud-hosted and still deliver poor calls, because the thing that determines call quality is the path the packets take, and most cloud VoIP providers do not control that path.
The three numbers that decide call quality
Voice is unusual among internet traffic: it is small, constant, and utterly intolerant of delay. Three measurements matter.
Latency, how long a packet takes to arrive. Under about 150ms one way, a conversation feels natural. Past 250ms people start talking over each other and the call feels like a satellite link. Latency is mostly a function of physical distance and the number of hops.
Jitter, variation in that delay. Packets arriving at inconsistent intervals are worse than packets arriving consistently late. The receiving device holds a jitter buffer to smooth it out, which itself adds delay; heavy jitter means a bigger buffer, which means more latency. Above roughly 30ms it becomes audible.
Packet loss, packets that never arrive. Voice has no time to retransmit, so loss is simply gone audio. A little is masked by concealment algorithms. Past a few percent you get the clipped, underwater quality everyone recognises.
The reason to care about all three rather than just the first is that they trade off against each other. A provider can make latency look good in a test while jitter quietly ruins real conversations.
Where “Tier-1 transit” comes in
Internet networks are loosely tiered. A Tier-1 network can reach every other network on the internet without paying for transit, it peers rather than buys. Below that, networks buy transit from someone larger.
Buying Tier-1 IP transit means your traffic gets onto a well-connected backbone quickly, rather than traversing several smaller networks first. In practice it means:
- Fewer hops, so less accumulated latency and fewer queues to sit in
- Better-provisioned links, so less congestion at peak
- More direct paths, so a call between two US cities stays on domestic infrastructure
It is not magic and it is not a guarantee. It is a meaningful reduction in the number of places things can go wrong.
Why “cloud VoIP” isn’t automatically any of this
A cloud VoIP provider runs its software in a cloud region. That says where the servers are. It says nothing about how the voice packets get from a caller to those servers, or from those servers to the destination carrier.
Two specific problems:
Trombone routing. If the platform is hosted in one region and your callers and destinations are in another, media can travel to the cloud region and back. A local call becomes an intercontinental round trip. It works. It sounds like it should not.
Best-effort public internet on the carrier leg. The cloud region has excellent connectivity. The hop from that region to the terminating carrier in your destination country may be entirely ordinary. Quality is set by the worst link, not the best.
This is why “we run on a major cloud provider” is not an answer to “what is the latency from Dallas to a Verizon mobile in Chicago”.
Regional gateways, and why they help
The fix for a lot of this is unglamorous: put equipment near the traffic.
If SIP termination homes to a US carrier near the traffic, the call reaches the terminating network quickly and stays there. The internet portion of the journey is short and the carrier portion is domestic, rather than crossing the country twice before it lands.
This is also why coverage claims are worth probing. “Nationwide coverage” might mean direct interconnects with the major US carriers, or it might mean a wholesaler somewhere will carry the call. Both are true statements. They sound different on the phone.
Measuring it yourself
Do not take anyone’s word for it, including ours.
- Place real calls at your busy hour, from where your callers actually are, to where you actually call. A quiet Tuesday morning test proves nothing.
- Look at jitter and loss, not just MOS. A single averaged score hides the variance that ruins individual calls.
- Check the whole path. Your own LAN and internet connection are part of it. Voice sharing an uncontended link with backups will suffer regardless of the carrier.
- Test over days, not minutes. Congestion problems are time-of-day problems.
If you want a trunk to run those tests against before committing to anything, ask us, we would rather you measured it than believed us.