The market divides cleanly once you apply the distinction from the last chapter but one: queue managers that hand each message to a consumer and forget it, logs that remember, and managed cloud services that sell either one by the request. Ownership has also changed more in the last four years than in the previous twenty, so the table records who owns what as of September 2026.

Two ownership facts to have ready. Cloud Software Group is the company formed when the private-equity firms Vista Equity Partners and Evergreen Coast Capital completed their purchase of Citrix on 30 September 2022, valued at $16.5 billion, and combined it with TIBCO. And IBM now owns IBM MQ, Red Hat’s messaging products and Confluent — more of a bank’s messaging estate than any vendor has before.

Every messaging capacity plan starts from the same four inputs: messages per second at peak, message size, whether messages are persistent, and how long they must be kept — a retention period for a log, a tolerated outage for a queue.

Kafka’s tiered storage, which moves older data to object storage so broker disks hold only the recent part of the log, changes the disk line dramatically; see streaming internals.

Within a site or across nearby zones, the majority rule above gives no data loss. Across distant regions it cannot, because waiting for a far-away majority on every write is too slow. Every product therefore replicates to a second region asynchronously — IBM MQ’s Cross-Region Replication for Native HA, disaster-recovery configurations of RDQM (the Replicated Data Queue Manager), Kafka’s MirrorMaker 2 or Confluent’s cluster linking, RabbitMQ’s federation and shovel plug-ins — and accepts that the last few seconds of messages may be missing after a regional failover. The recovery point objective for regional loss is therefore never zero, and the reconciliation process in the flows chapter is what closes the gap.