As a solutions architect designing a hybrid architecture, how should you extend an on-premises infrastructure to AWS to ensure high availability, low latency, cost minimization, and acceptable performance during primary connection failure?