Dedicated vs Shared Internet: Committed Rate or Best Effort

The term that separates the two products is the committed information rate. MEF 10.4, the industry specification for subscriber Ethernet services, describes a subscriber's traffic in terms of a committed rate, the CIR, and an excess rate, the EIR: frames within the committed rate are service-assured and the service-level specification applies to them, frames in the excess band are delivered when capacity allows and are exempt from it, and frames beyond both are dropped. A dedicated internet circuit is sold on that model. The rate on the order is a CIR, and the SLA's latency, jitter, loss and availability figures are promises about traffic inside it.

A shared plan has no CIR. The advertised number is a ceiling the access segment can deliver when nobody else is using it, and the carrier's own pages say so: performance is "up to" the advertised speed and depends on local network usage. That is not a defect; it is the pricing model. A cable node or a PON split shares its capacity among every subscriber on it, and selling a ceiling rather than a floor is what makes a shared plan cost a fraction of a circuit.

Every access technology except a dedicated circuit shares its last segment. On cable, every modem on a coax node shares that node's downstream and upstream spectrum; on a passive optical network, every optical network terminal on a splitter shares one port on the carrier's line terminal; on fixed wireless, every subscriber on a sector shares the sector. The ratio of subscribers to the capacity they share is the contention ratio, and it is set by the operator, not printed on the plan. When the neighbors are busy, the plan delivers less than its ceiling, and the operator has committed to nothing that says otherwise.

A dedicated circuit removes the neighbors. The bandwidth on that physical connection belongs to one customer, so the only contention is the customer's own traffic, which the customer can see, shape and plan for. That is why a dedicated circuit holds its rate at the same level regardless of the time of day, and why the carrier can put numbers on latency and jitter: it controls the whole path from the customer's demarc to its core.

An SLA is a list of measurements and the credit the carrier owes when it misses them. The usual entries are availability, stated as a percentage of the month the circuit is up; mean time to repair, the clock that starts when a trouble ticket is opened; and the network performance figures, round-trip latency, jitter and packet loss, measured across the carrier's network rather than to an arbitrary internet host. The remedy is a bill credit, and it is claimed, not automatic, so the customer's own monitoring is what proves a miss. Read the measurement points before the percentages: an SLA measured core-to-core says nothing about the last mile unless the last mile is named.

The other half of the agreement is what the carrier watches. A dedicated circuit is normally monitored by the carrier down to the edge device it installs, with alarms raised on its side before a user calls. On a shared plan the operator watches the node; the customer's link is invisible to it until a ticket is opened. For a site that carries an uptime clause to a customer of its own, that difference is the reason the circuit exists.

A site that hosts anything reachable from the internet, a VPN concentrator, a mail relay, a customer portal, a camera system with remote access, needs addresses that do not change. Static public addresses are the normal state of a dedicated circuit, and a block or a BGP session can be ordered when the site announces its own space; on a shared plan a static address, or with some operators a small block of them, is an add-on where the operator offers one and a block is unusual. The AT&T and Verizon business pages both draw the line here: static addressing and on-site hosting are listed among the reasons to move to a dedicated product.

The circuit is not the whole design. A dedicated circuit and a shared plan from a different carrier on a different medium, or an LTE/5G failover router, cover each other's failure modes at a fraction of a second circuit's cost, and an SD-WAN or dual-WAN edge decides which traffic rides which link. The circuit is quoted per address on the connectivity pages; the edge router, the firewall it lands on and the failover path are equipment we quote and stage, and they are the picks below.