The primary consideration is that the participating hosts must support dual stack. The pilot is contained within a single VLAN, so the application traffic remains inside the same Layer 2 broadcast domain across the dual-site data center environment. Layer 2 switches forward Ethernet frames and do not need to understand the IPv6 header to carry IPv6 traffic inside the VLAN. Layer 3 switches need IPv6 capability only where IPv6 routing, gateways, ACLs, or services are required. For this pilot, the essential requirement is that the servers and endpoints running the IPv6 application can operate with both IPv4 and IPv6 enabled while the existing environment continues to support IPv4. Dual stack allows IPv6 testing without removing IPv4 reachability or forcing translation. Requiring every Layer 2 and Layer 3 switch in both data centers to be dual-stack overstates the design requirement. The hosts participating in the pilot are the critical dependency. Reference topics: IPv6 migration, dual stack, Layer 2 VLAN transport, data-center pilot design, IPv4 and IPv6 coexistence.

An out-of-band management network should be isolated from the production data network and protected with strict access control. Cisco SAFE design principles treat the management plane as a sensitive function because it provides administrative access to infrastructure devices. Isolation ensures that production outages, routing problems, or compromised user segments do not automatically remove the operator’s ability to manage routers, switches, firewalls, and controllers. Access control ensures that only authorized administrators and management systems can reach the management interfaces and protocols. The management network should not be treated as a backup data path, because that undermines separation and can expose management devices to production traffic risks. It should also not be used as a general-purpose data backup network. “Facilitate network integration” is too vague and can conflict with the isolation requirement. Therefore, the best practices are to enforce access control and ensure network isolation. In a mature design, this normally includes dedicated management interfaces or VRFs, jump hosts, AAA, logging, and tightly controlled management protocol access.

Graceful restart is the correct high-availability design when the routing process on an intermediate OSPF router must undergo maintenance while the data plane remains available. Cisco describes graceful restart as a mechanism that allows a restarting router to continue forwarding traffic while neighbors temporarily preserve adjacency and route information. In OSPF, helper-capable neighbors assist the restarting router by maintaining forwarding state during the restart interval, preventing unnecessary route withdrawal and reconvergence. BFD is a failure-detection mechanism, so it would actually accelerate detection of a failure rather than preserve the forwarding path during planned process maintenance. Nonstop forwarding must be enabled on the device performing the control-plane restart and supported by neighbors, but the broad answer choice that captures the required routing-protocol behavior across the design is graceful restart on the participating routers. Configuring it only on R1 and R3 or only on R3 would not support the R2 process maintenance scenario. Reference topics: OSPF graceful restart, NSF, helper mode, control-plane maintenance, data-plane availability.

BGP community is the correct attribute because it can support both policy requirements through coordination with the service provider. Communities are route tags that an upstream provider can match to apply routing policy, such as adjusting local preference to influence which dual-homed provider link is preferred for inbound traffic. Communities also include well-known values used to constrain route propagation. Cisco documents no-export as a community that prevents a route from being advertised outside the receiving AS, and no-advertise as a community that prevents advertisement to any peer. This directly matches the requirement that several customer-advertised networks must propagate no further. MED can influence inbound path selection in some dual-link designs, but it does not provide the propagation-control function. AS-path prepending can influence inbound traffic but is less precise and again does not solve the no-further-advertisement requirement. Local preference is normally applied inside an AS, not directly by the customer to external peers. Reference topics: BGP communities, no-export, no-advertise, provider policy, inbound traffic engineering.

Implementing direct Internet access at every Cisco SD-WAN site improves application experience and reduces unnecessary backhaul. Many modern business applications are hosted in SaaS or public cloud environments. If branch Internet traffic must hairpin through a central data center before reaching those applications, latency increases and central WAN or security stacks can become bottlenecks. SD-WAN DIA allows selected cloud and Internet-bound traffic to exit locally while still applying policy, segmentation, and security controls. This decreases latency to applications hosted by public cloud providers and improves user experience. It also alleviates traffic on MPLS circuits because Internet and cloud traffic no longer needs to consume private WAN bandwidth back to a centralized egress point. DIA does not inherently reduce the physical latency of an Internet circuit, and it does not increase the provisioned circuit bandwidth. Zero-touch provisioning may accelerate site deployment, but that is a separate SD-WAN onboarding benefit rather than a specific benefit of deploying DIA at every site. Therefore, the two correct benefits are reduced latency to public cloud applications and reduced load on MPLS circuits.

For in-band management over the production WAN, the proper marking is DSCP CS2 mapped into the management or operations class with a minimum bandwidth guarantee. Cisco enterprise QoS guidance separates network control traffic from network management traffic. Routing protocol and control-plane traffic should not be diluted by ordinary management flows. CS6 is typically reserved for routing and network-control traffic; EF is reserved for low-latency voice bearer traffic; CS1 is commonly used for scavenger or low-priority traffic. Network management flows such as SSH, SNMP, NTP, FTP-based maintenance, and syslog require reliable delivery, but they should not starve business traffic or real-time services. Therefore, marking with CS2 and mapping the traffic into Class2 is the correct design choice. The bandwidth adjustment should come from a lower-priority class, not from the routing/control or real-time class. Because this is in-band management, the architect must also ensure access control, source restrictions, AAA, logging, and management-plane protection. QoS classification alone does not secure management access; it only gives the traffic an appropriate forwarding treatment over the MPLS WAN. Reference topics: in-band management, DSCP CS2, enterprise QoS classes, management-plane traffic.

Bandwidth percentage with policing is the correct QoS approach. The requirement has two separate parts: each subnet must receive a minimum share during congestion, and no subnet should introduce delay into download traffic. Bandwidth percentage is used to allocate a minimum bandwidth guarantee among classes during congestion while still allowing a class to use unused bandwidth when other classes are idle. Policing enforces a rate by dropping or remarking excess traffic rather than buffering it. That behavior is important because the question explicitly says download traffic must never experience delay. Shaping would be wrong because shaping buffers excess packets and transmits them later, which intentionally adds delay. Rate- limiting is a broad term, but when the design choice offers policing explicitly, policing is the precise mechanism that matches the no-delay requirement. A parent or class policy can reserve 25 percent per subnet on a 40 Mbps service, giving each subnet 10 Mbps during congestion while unused bandwidth remains available. Reference topics: CBWFQ bandwidth percent, traffic policing, shaping versus policing, Internet edge QoS, congestion management.

Single-topology IS-IS with the transition feature is the correct design when IPv6 is being added to an existing IPv4 IS-IS network and both topologies must match exactly. In single-topology mode, IS-IS calculates one topology and uses it for both IPv4 and IPv6, which fits the requirement that the IPv4 and IPv6 paths and router levels remain the same on each interface. Cisco documentation notes that single-topology IS-IS IPv6 requires routers to run the same set of address families for adjacency consistency; during migration, transition support allows the network to operate while IPv6 is introduced gradually. Multi-topology IS-IS is used when IPv4 and IPv6 should be allowed to follow different topologies or when not all routers support the same address families. The question explicitly says the topologies will match exactly, so multi-topology is unnecessary. Single topology without the transition feature is risky during phased migration because routers that do not yet run both address families may fail adjacency checks or create blackholing. Reference topics: IS-IS for IPv6, single-topology IS-IS, transition mode, address-family consistency, IPv6 migration.

BGP outbound route filtering is the correct solution because the enterprise wants control over which routes it receives without waiting for a provider change cycle each time the prefix set changes. Cisco BGP prefix-based ORF allows a receiving BGP speaker to send an inbound prefix-list filter to the remote peer, which then applies that list as an outbound filter. This reduces unwanted routing updates before they cross the session and lets the enterprise update the desired prefixes locally when business requirements change. Requesting only fixed prefixes from the ISP would work initially but fails the flexibility requirement because the provider has a two-week change process. Requesting a full table and filtering inbound would provide flexibility but wastes memory and CPU on the customer router because all routes are still received before filtering. A maximum-prefix limit does not select the correct routes. ORF is therefore the scalable operational design for receiving a default route and selected prefixes with minimal provider involvement. Reference topics: BGP ORF, prefix-based filtering, inbound route control, ISP peering design, routing table scale.

The correct BGP design is to attach the well-known No-Export community to the prefixes that should not be advertised beyond the service provider boundary. Cisco BGP communities are policy attributes that allow routes to carry instructions or classifications across BGP peers. The No-Export community tells a receiving BGP speaker not to advertise the route outside its confederation or autonomous system. That behavior is exactly what is needed when branch prefixes may be exchanged with the provider but must not be propagated to the public Internet. The No-Advertise community is stronger: it prevents the route from being advertised to any BGP peer, which may block reachability even where the provider still needs to know the route. NOPEER is used for route-server or peering policy control and is not the clean answer for excluding branch prefixes from Internet propagation. Setting a generic “Internet community” is not a standard solution by itself. The architect should also ensure that outbound prefix filters and provider-side community support are documented in the service agreement. Reference topics: BGP communities, No-Export, No-Advertise, provider edge filtering, Internet route policy.