Integrated Services
/ IntServ: INT-serv /
Think of how you might guarantee a good meal for a big group: call ahead and reserve a table for exactly your party, so the restaurant sets aside the seats and prepares for you. Integrated Services (IntServ) is that reservation idea applied to the network. Before sending a real-time flow, an application asks the network to reserve resources — a specific amount of bandwidth and a delay bound — along the entire path from sender to receiver, and the network either grants the reservation (admitting the flow) or refuses it (so it cannot start and ruin others). It promises per-flow quality, not just per-class favoritism.
It works through a signaling protocol called RSVP (the Resource Reservation Protocol). The sender describes its traffic (often as a token-bucket profile: average rate and burst), and the receiver sends an RSVP reservation request back along the path. Every router on the way runs admission control: it checks whether it still has enough spare capacity to honor this flow's request on top of all its existing reservations. If every router along the path says yes, the reservation is installed hop by hop and the flow gets a guarantee; if any one router cannot fit it, the reservation is rejected. Each router must then keep per-flow state and schedule accordingly (for example with weighted fair queuing) to actually deliver what it promised.
Why it matters: IntServ can offer the strongest guarantee — a true per-flow promise of bandwidth and bounded delay — which is exactly what a sensitive video call would want. But it carries a heavy honest cost: every router on the path must remember and police the state of every individual flow. On a backbone router carrying millions of simultaneous flows, that per-flow state and processing simply does not scale. This is precisely why the Internet largely turned away from IntServ toward the coarser, stateless-in-the-core Differentiated Services model, and why IntServ today is found mostly within smaller, controlled networks rather than across the global Internet.
A video app wants 4 Mbps with a tight delay bound. Via RSVP, each router on the path checks its free capacity; routers 1 through 4 can fit it, so they install the reservation and the call is guaranteed. Had router 3 been full, the reservation would have been refused and the app told it cannot get the guarantee.
IntServ reserves per-flow resources hop by hop via RSVP — a guarantee, but stateful at every router.
IntServ's fatal scaling problem is per-flow state: a core router would have to track and schedule millions of individual reservations. That is why the Internet leaned toward DiffServ, which keeps the core stateless by working on traffic classes instead of flows.