Cross-AZ Latency Patterns

Sub-ms typical.

Overview

Cross-AZ latency in modern AWS regions is typically sub-millisecond, but it is not zero, and the bandwidth between AZs is paid where intra-AZ traffic is often free. The discipline is to design with cross-AZ latency in mind: co-locate latency-sensitive stateful pairs (database and primary writer) in the same AZ, replicate stateless services across AZs for resilience, and monitor cross-AZ bandwidth as a cost line item.

• Sub-millisecond typical. Modern AWS regions have AZs close enough for sub-ms latency under normal conditions.
• Higher than intra-AZ. Same-AZ latency is lower; latency-sensitive write paths benefit from co-location.
• Lower than cross-region. Cross-region latency is tens of milliseconds; the cross-AZ vs cross-region tradeoff is significant.
• Bandwidth costs plus variance. Cross-AZ traffic is billed; cross-AZ p99 can spike during AZ-level congestion, affecting tail latency.

The approach

The practical approach is to co-locate database and primary writer in the same AZ for write-path latency, replicate stateless services across AZs for resilience, use AZ-aware load balancing where the LB supports same-AZ preference, monitor cross-AZ bandwidth as a cost line item, and document per-service AZ placement so the topology is reviewable.

• Co-locate database and primary writer. Same AZ keeps writes fast; the latency tax of cross-AZ writes adds up at scale.
• Replicate stateless services. Stateless workloads spread across AZs; the load balancer absorbs single-AZ failure cleanly.
• AZ-aware load balancing. Some load balancers support same-AZ preference; reduces cross-AZ chatter for latency-sensitive traffic.
• Monitor cross-AZ traffic plus documented topology. Cross-AZ bandwidth tracked as a cost line; per-service AZ placement committed for operational review.

Why this compounds

Cross-AZ awareness compounds across services. Each correct placement preserves write-path latency; each AZ-aware load balancing decision saves cross-AZ bandwidth; the team builds intuition for AZ topology that pays off on every new service.

• Latency. Right placement for latency-sensitive workloads; the user sees sub-millisecond write paths rather than cross-AZ tax.
• Cost. Less cross-AZ traffic; the bill tracks the actual topology rather than accidental routing.
• Resilience. Multi-AZ replication where it matters; stateless services survive AZ failure without manual intervention.
• Institutional knowledge. Each placement decision teaches AWS networking; the team learns where co-location pays off and where replication matters.

Cross-AZ latency awareness is an operational discipline that pays off across years. Nova AI Ops integrates with cross-AZ telemetry, surfaces topology patterns, and supports the team’s network engineering discipline.