Sydney Says Hello

Users in Sydney hitting an application deployed only in Virginia see about 230 ms per request. Tracing shows 15 ms of server-side processing and the rest spent on the network. The team proposes a week of profiling to speed the handler up by 5 ms. What is the right read of the situation?

The transport is the problem, so moving from HTTP/1.1 to HTTP/3 with 0-RTT resumption removes the propagation component of the delay rather than only the handshake round trips.

The time is dominated by propagation across a long path, bounded below by light in fiber at about two thirds of c, so the wins come from serving closer to Sydney and cutting round trips.

The added time is queueing at the Virginia instances under load arriving from distant regions, so provisioning more application servers there brings 230 ms down toward the 15 ms of processing.

The path between the regions is bandwidth constrained at peak, so buying a higher-capacity link between Sydney and Virginia lowers the round trip time in proportion to the capacity added.