Refer to the exhibit. After this new OSPF design with per-packet load balancing was implemented, Host A reported that large file downloads from Server A became slow and sometimes failed. The operation

Upgrading all links to 10 Gbps is not cost-conscious and does not inherently eliminate equal-cost paths; it increases bandwidth but does not resolve the per-packet reordering caused by ECMP.

IP SLA probes provide monitoring and can trigger tracked route changes but do not alter OSPF cost calculations or remove equal-cost paths from the routing table.

Changing the reference bandwidth only on R4 creates asymmetric cost calculations between R4 and the rest of the OSPF domain, which introduces routing inconsistencies rather than resolving them.

OSPF calculates link cost as reference-bandwidth divided by interface bandwidth. When different routers use different reference-bandwidth values, they derive different costs for the same links, causing inconsistent ECMP path selection across the topology. Adjusting the auto-cost reference bandwidth uniformly on all routers ensures every device computes identical path costs, which can eliminate equal-cost paths and thereby disable per-packet load balancing - the root cause of the out-of-order packet delivery degrading TCP throughput.