CONCEPT Cited by 2 sources
Cross-AZ replication bandwidth cost¶
Definition¶
Cross-AZ replication bandwidth cost is the per-byte cloud- provider charge for transferring data between two availability zones within the same region — the unavoidable cost of multi-AZ replication on any durable streaming workload running on AWS / Azure / GCP.
Unlike intra-AZ transfer (free on every major cloud) and
cross-region transfer
(priced on a separate meter, typically several cents per GB),
cross-AZ transfer sits in an economic middle ground: it's not
free, but billing rates are lower than cross-region. Because
multi-AZ is the default high-availability posture (RF=3 with AZ
spread is the standard Kafka / Redpanda deployment shape), every
produced byte on a durable topic is replicated across AZs at least
RF-1 times, and each cross-AZ replica is a billed transfer.
On a stable-state production streaming cluster, cross-AZ replication cost is often a larger line item than the compute cost of the brokers themselves — the dominant direct-operational cost of running the cluster.
Canonical wiki source¶
Introduced in the Redpanda 25.3 launch post as the load-bearing cost axis motivating Cloud Topics:
"With Cloud Topics, each batch of messages is passed straight through and written to cost-effective object storage (S3/ADLS/GCS) while topic metadata is managed in-broker — replicated via Raft for high availability …"
"This approach virtually eliminates the cross-AZ network traffic associated with data replication. You keep millisecond performance where it matters, and pay object-store prices for replication where it doesn't, without sacrificing durability."
Concrete business-impact framing:
"Dramatically lower TCO: Sidestep steep cloud provider networking charges for compliance, security, training data, or batch analytics."
Why object storage sidesteps it¶
Cloud providers' multi-AZ durability guarantee on their object storage services (S3, ADLS, GCS) is already multi-AZ-replicated at storage-service cost — the provider amortises cross-AZ replication across all storage-service customers and doesn't bill it as a separate per-byte transfer.
Writing streaming records straight to object storage therefore shifts replication cost from:
- Cross-AZ egress (billed at per-GB rates on the broker's compute VMs)
to:
- Object-storage PUT / GET operations (billed at per-request rates, with multi-AZ durability included in the storage tier pricing)
The economics flip: for high-volume, low-read-rate streams (observability, compliance logs, model-training data) the object-storage shape is dramatically cheaper. For low-volume, high-read-rate streams the traditional NVMe + cross-AZ- replication shape is cheaper (and lower-latency).
The per-topic choice¶
The cost structure is workload-dependent — which is why Cloud Topics offers the choice per topic rather than per cluster. Two topics in the same cluster can have different storage substrates, different replication costs, and different latency profiles, aligned with the latency-criticality of the data they carry.
Position vs cross-region bandwidth cost¶
Cross-AZ and cross-region are two points on the same bandwidth- cost axis:
| Transfer | Rate | Latency | Dominant driver |
|---|---|---|---|
| Intra-AZ | Free | sub-ms | Free tier |
| Cross-AZ within region | $0.01-$0.02/GB | 1-5 ms | Multi-AZ HA replication |
| Cross-region | $0.02-$0.09/GB | 30-100+ ms | Multi-region DR / global latency |
Cross-AZ is the cost axis every multi-AZ streaming cluster pays. Cross-region is on top of that when you run a stretch cluster or MM2 / Shadowing between regions.
Cloud Topics attacks the cross-AZ axis specifically — Shadowing addresses the cross-region axis.
Seen in¶
- sources/2025-11-06-redpanda-253-delivers-near-instant-disaster-recovery-and-more — canonical wiki source. Redpanda 25.3's Cloud Topics feature is positioned explicitly as a cross-AZ-replication-cost elimination mechanism for latency-tolerant streaming workloads.
- sources/2026-03-30-redpanda-under-the-hood-redpanda-cloud-topics-architecture — deep-dive explaining how Cloud Topics attacks this cost axis: "Since Cloud Topics writes data directly to object storage, we can bypass the Cross-AZ networking tax incurred when replicating via Raft." Keeps metadata on the Raft path (a tiny fraction of produced bytes) via a placeholder batch while payload bytes go to S3/GCS/ADLS.
Related¶
- systems/redpanda — the broker hosting cost-tiered topics.
- systems/redpanda-cloud-topics — the feature that eliminates this cost per-topic.
- systems/aws-s3, systems/google-cloud-storage — the object-storage substrates that absorb the replication cost at storage-service pricing.
- concepts/cross-region-bandwidth-cost — the peer axis at the region-pair granularity.
- concepts/tiered-storage-as-primary-fallback — the prior Redpanda tiering model where cross-AZ cost is paid on the hot tier (NVMe) and avoided only on the cold tier.
- concepts/stateless-compute — the WarpStream-style shape Cloud Topics approximates per-topic.
- concepts/placeholder-batch-metadata-in-raft — the mechanism that lets Cloud Topics keep metadata on the Raft path while moving payload off it.
- concepts/l0-l1-file-compaction-for-object-store-streaming — the on-object-storage layout that replaces cross-AZ replication.
- patterns/per-topic-storage-tier-within-one-cluster — the Cloud Topics pattern that exploits this cost axis.
- patterns/tiered-storage-to-object-store — the broader pattern family.
- patterns/object-store-batched-write-with-raft-metadata — the canonical write-path pattern that attacks this axis.