On Mon, Sep 15, 2025 at 7:07 AM Mike Hammett via NANOG <nanog@lists.nanog.org> wrote:
*nods* Well, and that's the rub. Their expectations don't match any Internet SLA I've ever seen
The implied expectation is on a completely infeasible ground for the provider of a basic internet line. Typically security updates alone necessary for CPEs, etc, would bump connections more than a second per 6 months. It's more like the service level for point-to-point path-protected circuits, or dual-connection disparate-path dark fiber build directly between two locations, bought from a telecoms provider, and not an internet connection, in order to reasonably offer a service level anywhere near this. "1 second maximum every 6 months"; let alone less. An infrequent 1 second one-off interruption typically doesn't count as an outage. It may also be something that cannot be diagnosed without major maintenance; that is part of the environment IP exists in, and the internet itself is a large network with multiple instable or changing paths through other providers' networks. Most peers are a best-effort packet delivery, not a promise of 0% loss. As described the expectation the expectation is a near equal to guaranteeing a lossless connection between point A and B. But, point B is across the internet. Which means part of the path to B is outside the control of the provider of that line in the first place, and parts of that path at different points are on infrastructure which is shared and overcommitted both by the point A's network providers and point B's network providers. Also, point A and point B are both host devices which are subject to the chance of a local software or hardware issue. That means when there is a "disruption"; the rational thing for Provider A to do is to assume the issue is with point B, the network in the middle, or end devices, until proven otherwise. Providers do not assume an issue they have not detected is with their own service or network; that has to be proven, and the proof may be too difficult to accomplish. Provider B or the networks in the middle have no reason to adjust their service to accommodate Provider A or Provider A's end users' special requirements without first making purchases of additional dedicated infrastructure and contracts with each provider between point A and B. For a network line provider to ensure this level of service; that provider realistically has have to have failover circuits between point A and B with dedicated infrastructure on the whole path not dependent on 3rd parties. For example a ring path-protected point to point Ethernet or SONET-based circuits with established bandwidth reservations between A and B. Even with those: you can still expect a few hours of outage per year for maintenance. And there is always the chance of that one-off double fibre cut every few years and similar. The risk level depends on many variables. There are those SDN solutions that overlay private networks with redundant forwarding on top of several internet connections. _Both_ point A and point B require multiple internet connections. The multiple internet connections still have simultaneous failure scenarios during major internet events. The expectation can be more protection from different possible causes of outages, but the risk for them does not go to 0.0%, etc
Mike Hammett -- -JA