A regional hospital campus supports 3,000 endpoints with voice, video, and data sharing a converged core, and ICU phones need low latency during congestion without tracking every call across the network. The design must scale simply as new clinics are added next year. Which QoS model should the designer select to meet the scalability requirement?

DiffServ with edge classification and per-hop queuing fits because voice is marked once at the trust boundary with DSCP EF and each hop provides low-latency queuing without flow signaling, so adding clinics adds only edge policy while the core remains stateless and ICU voice retains delay priority during congestion. The IntServ RSVP guaranteed-delay approach fails scalability because it installs and refreshes reservation state along the path for admitted sessions, so router memory and signaling grow with calls and endpoints rather than staying flat across a 3000 endpoint campus. The DiffServ approach favoring video and data fails priority because per-hop behavior follows marking, so when queues build the scheduler serves the wrong aggregate first and ICU voice still experiences queuing delay and loss. The overprovisioning-only approach fails congestion control because added bandwidth lowers average utilization but offers no scheduling preference when bursts temporarily oversubscribe links, leaving voice vulnerable to jitter exactly when protection is required. Exam caveat: focus on constraints over benefits, as guaranteed service and abundant bandwidth sound sufficient until matched against without per-call tracking and low latency during congestion. Practical check: verify EF marking preservation, trust boundaries, and LLQ drops with show policy-map interface and captures during load testing before clinic expansion.