You migrated a monolith to GKE microservices that depend on on-prem CRM and MySQL (PII must stay on-prem). A Cloud VPN connects on-prem to Google Cloud, but fluctuating bandwidth causes latency spikes and failing requests. How do you address the latency caused by limited VPN throughput?

Increasing the number of Cloud VPN tunnels improves throughput by allowing more data to be transmitted concurrently, directly addressing the limited bandwidth causing latency spikes. While caching PII in Memorystore could reduce some on-prem calls, the problem states PII must stay on-prem, making this an unsuitable solution. Extending a service mesh with Istio enhances traffic management and observability but doesn't inherently increase the underlying network throughput. Reducing the MTU (Maximum Transmission Unit) would decrease packet size, potentially increasing overhead and further reducing effective throughput, rather than improving it.