Migrating Ctrip Hotel Direct Connect Service to AWS: Cloud‑Native Architecture, Cost and Performance Optimizations

Ctrip’s overseas hotel business experienced rapid growth in data transfer between global suppliers and its headquarters IDC, creating bandwidth, latency, and cost pressures on cross‑border network links.

The hotel direct‑connect system, which aggregates static, dynamic, and order data via automated interfaces, was originally deployed on Ctrip’s IDC and suffered from high bandwidth consumption because every request was proxied and full payloads were retrieved before processing.

To address these issues, Ctrip evaluated major public cloud providers and selected AWS for its global presence, network capabilities, and mature services. The migration plan involved moving the hotel‑direct service to AWS, using Amazon EKS as the managed Kubernetes platform.

Key cloud‑native redesign steps included moving supplier‑access modules to AWS, where request splitting and response filtering are performed, reducing payload size to 30‑40% of the original bandwidth usage.

A custom Squid proxy deployed on EKS handles outbound request traffic, allowing inbound response traffic (which is larger) to avoid NAT‑gateway charges, significantly lowering network costs.

By deploying services in AWS regions close to suppliers, network latency was reduced by up to 50% for distant providers, and cross‑AZ traffic was minimized using AWS NLB with disabled cross‑AZ routing and local traffic policies.

Elastic scaling was achieved with Kubernetes HPA and Cluster Autoscaler, combined with AWS Auto Scaling Groups and spot instances, cutting compute costs by 50‑80% while maintaining high availability.

DNS performance was improved by deploying NodeLocal DNSCache via a DaemonSet, lowering peak query latency from 2.5 seconds to 300‑400 ms, an 80% reduction.

Future work includes consolidating persistent storage back to the IDC, linking multiple AWS VPCs to reuse existing monitoring and logging frameworks, and further simplifying multi‑cloud operations.

The migration demonstrates how cloud‑native practices—container orchestration, spot instance usage, DNS caching, and cross‑AZ traffic control—can deliver substantial performance gains and cost savings for large‑scale, latency‑sensitive services.