The True Cost of Kafka Replication

Introduction

Kafka cluster-to-cluster data replication is critical to many use cases: disaster recovery (DR), cloud or data center migration, testing applications with production-like data, and multi-region data distribution. 

Easy replication of data between clusters:

• Increases streaming data availability across IT environments. 

• Accelerates migrations.

• Promotes data sharing inside and outside a business.

• Improves quality and speed of streaming application development.

The business case is clear, but the cost model is not. Some solutions appear free but impose heavy operational burden. And while commercial services remove some of that effort, they charge in usage-based models that can scale unpredictably and have you vendor-locked. 

With reducing infrastructure spend being a top concern for every business, this paper looks at the true cost of replication across four approaches – MirrorMaker 2, Confluent Replicator, Confluent Cluster Linking, and Amazon MSK Replicator – using two common scenarios (15 GB/day and 100 GB/day), before contrasting them with the fixed-price approach of Lenses K2K.

MirrorMaker 2: The Free but Costly Baseline

MirrorMaker 2 (MM2) is the open-source baseline, but running it is far from free. MM2 requires teams running Kafka Connect clusters, complex and expensive infrastructure to manage and scale (Confluent Docs). Dedicated Connect clusters are usually required to isolate replication from other tasks, which translates into more infrastructure spend and monitoring overhead (Confluent Replicator Tuning).

A major downside is the engineering complexity to deploy and manage. This can normally only be done by Kafka experts.  For example, offset translation for seamless consumer failover requires very careful tuning (KIP-716). Schema replication must be handled separately since MM2 does not migrate schemas. Any filtering or masking requires custom code in Single Message Transforms. Support is limited to community forums, with no official SLA.

In practice, MM2 doesn’t require vendor fees but trades them for engineering hours, infrastructure duplication, and accumulated technical debt.

Confluent Replicator

Confluent Replicator is Confluent’s enterprise-supported version of MirrorMaker 2. In Confluent Cloud it is offered as a managed Kafka connector, billed by task-hours plus throughput (pre-compression). According to Confluent’s price list, throughput is charged at $0.025 per GB transferred in U.S. regions, and each connector task is billed at $0.20 per hour (Confluent Connector Pricing). Optional add-ons such as PrivateLink connectivity add small hourly surcharges.

If higher throughput requires scaling up, costs rise linearly with the number of connector tasks. Replicator still runs on the Kafka Connect framework and provides only at-least-once semantics, meaning duplicates can occur. Its value lies in vendor support, monitoring integration via Confluent Control Center, and enterprise SLAs.

However, Replicator is not vendor-agnostic. It is a proprietary Confluent component, supported on Confluent Platform and Confluent Cloud. There may be issues with Confluent working with non-Confluent based Kafka clusters such as AWS MSK or Redpanda. This vendor tie-in limits flexibility and can increase lock-in costs if your architecture spans multiple Kafka flavors.

Confluent Cluster Linking

Cluster Linking takes a different approach: broker-native replication. In Confluent Cloud it is billed by $0.25 per hour per link per cluster, plus $0.25 per GB replicated (Confluent Cluster Linking Pricing). In APAC, multipliers can increase the hourly rate by ~20%.

Cluster Linking allows for exactly-once semantics, preserves offsets exactly and supports seamless failover. The drawback is still vendor lock-in – it is only supported between Confluent-managed clusters – and it does not support filtering or transforms.

Amazon MSK Replicator

Amazon MSK Replicator is AWS’s managed Mirror Maker 2 option, launched in 2023. Pricing is $0.30 per hour per replicator instance plus $0.08 per GB processed (AWS MSK Pricing). Cross-region transfer is billed separately at AWS egress rates.

MSK Replicator is simple to enable and fully managed, but it only works MSK-to-MSK within the same AWS account (AWS Docs) and comes with more limitations than what Mirror Maker 2 does. And, like MM2, it offers no filtering or schema sync.

Why Costs Escalate

Costs rise quickly as replication expands. Usage-based pricing doubles when traffic doubles, making spikes and backfills expensive. Multi-region setups multiply costs by adding links or replicators, and cross-region traffic adds significant egress charges.

Disaster recovery patterns often require mirroring entire clusters, doubling infrastructure. Engineering time also grows with scale: managing schema registries, offset syncs, and topology automation adds hidden TCO.

Even with managed services, the combination of usage fees, infra duplication, and human overhead leads to unpredictable and often rising bills.

Lenses K2K: Fixed and Predictable

Lenses K2K Replicator takes a different path. It is priced at a flat $200 fee per month per replicator instance, regardless of data throughput. This means costs scale linearly with the number of replication pipelines, not with the volume of events.

Operationally, K2K is Kubernetes-native and lightweight, avoiding Kafka Connect’s cost and complexity. It includes exactly-once delivery, schema registry sync, and filtering/masking/routing out of the box. It works across any Kafka flavor such as: Apache Kafka, Confluent, AWS MSK, and even Azure Event Hubs.

Comparison at a Glance

Conclusion

Simple and cost efficient replication is essential for a modern streaming data architecture. But costs vary dramatically. MirrorMaker 2 trades license savings for heavy engineering overhead. Confluent and AWS managed services reduce ops work but impose usage-based bills that scale with every gigabyte.

Lenses K2K takes a different approach. It delivers enterprise-grade replication, with advanced features including exactly-once semantics and schema migration. It is vendor-agnostic, working across Apache Kafka, Confluent, AWS MSK, Azure Event Hubs or other Kafka-compatible APIs to avoid vendor lock-in. 

Most importantly, K2K is cheaper and simpler to run. At a flat monthly fee per replicator instance, its cost remains predictable and independent of throughput, while Confluent and AWS bills rise with every spike in traffic or new link. Combined with its Kubernetes-native design and built-in features (schema sync, filtering, routing), K2K minimizes both infrastructure waste and engineering overhead.

As a final convenience, K2K can be integrated into Lenses DevX UI, enabling platform engineers to delegate self-service replication capabilities to engineering teams. This offers the same user-friendly interface and governance model that they use for their broader Kafka operations. Making replication not just more cost-effective and reliable, but also easier to manage day-to-day. 

In replication, predictability is often the most valuable cost saving of all. K2K provides both the technical guarantees and the operational convenience that modern data platforms need.

Want to see the difference Lenses K2K can make?

Use our Savings Calculator to calculate the savings with K2K for your setup.