Comparison of managed messaging and streaming solutions available on AWS platform.
| Parameter | SQS / SQS FIFO | SNS / SNS FIFO | EventBridge | Kinesis Data Streams | IoT Core Broker | MSK/Kafka (2.2.1-2.7.0) | Amazon MQ (RabbitMQ 3.8.6) | Amazon MQ (ActiveMQ 5.15 Classic) | ElastiCache (Redis 6.x) |
|---|---|---|---|---|---|---|---|---|---|
| Infrastructure | Serverless | Serverless | Serverless | Serverless | Serverless | Managed (EC2) | Managed (EC2) | Managed (EC2) | Managed (EC2) |
| HA | Regional | Regional | Regional | Regional | Regional | Multi-AZ | Multi-AZ | Multi-AZ | Multi-AZ |
| SLA | 99.90% | 99.90% | 99.99% | 99.90% | 99.90% | 99.90% | 99.90% | 99.90% | 99.90% |
| Order | Best-effort/FIFO | best-effort/FIFO(1) | best effort (no order) (3) | Ordered (within a shard) | N (9) | Y | Y | Y | Y (Streams) / N (Pub/Sub) |
| Delivery | At least once / exactly once (FIFO) | At least once / exactly once (FIFO) (4) | At least once | At least once | QoS0 (can drop message), QoS1 (at least once) (9) | At least once, At most once, Exactly once (55,56) | At least once, At most once (54) | Depends on protocol (exactly once is possible with MQTT/QoS2)(14) | At least once (Streams) / At most once (Pub/Sub) (15) |
| Routing | N | Y (via filtering) (62) | Y (63) | Y (b/w shards using partition key) (64) | Y (65) | N (66) | Y (67) | Y (with Apache Camel) (68, 69) | Y (Pub/Sub pattern matching) (70) / N (Streams) |
| Max message size | 256Kb | 256 Kb | 256 Kb | 1MB | 128 Kb | 1 MB (default) | 128Mb (in v3.8.0+)/2Gb (up to v3.7.0) (61) | some Gb (limited by RAM, disk size) (59, 60) | 32 Mb/min* (Pub/Sub) (75) / 512 Mb (Streams) (76) |
| Max rate, req/s | 3000msg/sec (batch mode) | 3000msg/sec or 10MB/sec (batch mode) | 2400 put msg/s; 4500 invoc/s (FRA, soft limit), | 1000 msg/sec/shard (max 1Mb/s) (24) | 100 req/sec, 512Kb/s (31) | nM msg/sec (11, 12) | 1M msg/sec (10) | 22K msg/sec (58) | 1M msg/sec (16) |
| Retention, max | 14 days | 6h (SMTP, SMS, mobile push) /23d (SQS, Lambda) (19) | 24h (5) | 1Y | 1h (retry/QoS1) (31), 7d (QoS1 persistent sessions, 42) | forever (limited by disk) | Limited by disk? (13) | forever (limited by disk) | forever (Streams, limited by disk) (72) / N (PubSub) (71) |
| Replay | N (2) | N (2) | Y (3) | Y | N | Y | N (25) | N (27) | Y (Streams) / N (Pub/Sub) |
| P2P vs Pub/Sub | P2P | Pub/Sub | Pub/Sub | Pub/Sub | Pub/Sub | Pub/Sub | Pub/Sub | Pub/Sub | * |
| Push vs Pull | Pull | Push | Push | Both (77) | Push | Pull | Both | Both | Pull (Streams) (74) / Pull (Pub/Sub) (73) |
| Encryption in transit | Y | Y | Y | Y | Y | Y | Y | Y | Y (18) |
| Encryption at rest | Y | Y | Y | Y | ? | Y | Y | Y | Y (17) |
| DLQ | Y | Y (7) | Y (23) | N (21, 22) | N | Y (via Connect/emulation) (30) | Y (28) | Y (26) | Y (29) |
| Message Schema (on AWS) | N | N | Y (33) | Y (with KPL/KCL) (34) | N | Y (32,35) | N | N | N |
| Standards, Protocols, Transports | JMS1.1, REST (36) | REST | REST | REST | REST, WebSockets, MQTT* (37) | Proprietary | AMQP, STOMP, MQTT, HTTP, WebSockets (38), JMS1.1 (39) | AMQP, AUTO, MQTT, OpenWire, REST, RSS/Atom, STOMP, WSIF, WS Notification, XMPP, JMS 1.1 (40,41,57) | Proprietary |
| Data type | Text, XML, JSON (46) | Text, JSON (45) | JSON (47) | Base64-ed blob (44) | JSON, Binary (43) | Binary (48) | * | * | Binary-safe K/V (49) |
| Processing Latency | 0.13-1.3sec (6,53) | seconds (51) | 0.5 sec (4) | 200ms-1s / 70ms (with enhanced fan-out) (50,77) | ? | 5-48ms (52,53) | 1-120ms (52,53) | 49ms (53) | 1ms (16) |
(1) https://aws.amazon.com/blogs/compute/building-event-driven-architectures-with-amazon-sns-fifo/ and https://aws.amazon.com/about-aws/whats-new/2020/10/amazon-sns-introduces-fifo-topics-with-strict-ordering-and-deduplication-of-messages/
(2) Replay implementation with SNS + 2 x SQS https://docs.aws.amazon.com/sns/latest/dg/sns-fork-pipeline-as-subscriber.html#sns-fork-event-replay-pipeline
(3) https://aws.amazon.com/blogs/aws/new-archive-and-replay-events-with-amazon-eventbridge/
(4) https://docs.aws.amazon.com/sns/latest/dg/fifo-message-dedup.html and SNS vs EventBridge https://cloudonaut.io/eventbridge-vs-sns/
(5) https://docs.aws.amazon.com/eventbridge/latest/userguide/create-eventbridge-scheduled-rule.html
(6) https://softwaremill.com/amazon-sqs-performance-latency/
(7) DLQ https://aws.amazon.com/about-aws/whats-new/2019/11/amazon-sns-adds-support-for-dead-letter-queues-dlq/ and https://aws.amazon.com/blogs/compute/designing-durable-serverless-apps-with-dlqs-for-amazon-sns-amazon-sqs-aws-lambda/
(8) https://aws.highspot.com/items/5f5fe9b260e9cc1120781e22?lfrm=srp.0#31 - comparison of different solutions
(9) https://docs.aws.amazon.com/iot/latest/developerguide/mqtt.html
(11) https://events.static.linuxfound.org/sites/events/files/slides/Kafka%20At%20Scale.pdf
(12) https://engineering.linkedin.com/kafka/benchmarking-apache-kafka-2-million-writes-second-three-cheap-machines and 15K msg/sec https://aiven.io/blog/benchmarking-kafka-write-throughput-performance-2019-update
(13) https://www.rabbitmq.com/ttl.html
(14) https://activemq.apache.org/components/artemis/documentation/2.3.0/protocols-interoperability.html
(15) https://redislabs.com/solutions/use-cases/messaging/
(16) https://devopedia.org/redis-streams
(17) https://docs.aws.amazon.com/AmazonElastiCache/latest/red-ug/at-rest-encryption.html
(18) https://docs.aws.amazon.com/AmazonElastiCache/latest/red-ug/in-transit-encryption.html
(19) https://docs.aws.amazon.com/sns/latest/dg/sns-dead-letter-queues.html
(20) https://docs.aws.amazon.com/amazon-mq/latest/developer-guide/data-protection.html
(22) https://acloudguru.com/blog/engineering/use-sns-to-retry-failed-kinesis-events
(23) https://docs.aws.amazon.com/eventbridge/latest/userguide/rule-dlq.html
(24) https://docs.aws.amazon.com/streams/latest/dev/service-sizes-and-limits.html
(25) https://www.cloudkarafka.com/blog/which-service-rabbitmq-vs-apache-kafka.html
(26) https://activemq.apache.org/message-redelivery-and-dlq-handling
(27) History can be emulated via Mirrored queues http://activemq.apache.org/mirrored-queues.html
(28) https://www.rabbitmq.com/dlx.html
(29) https://redis.io/topics/streams-intro
(30) https://www.confluent.io/blog/kafka-connect-deep-dive-error-handling-dead-letter-queues/ and https://towardsdatascience.com/dead-letter-queue-dlq-in-kafka-29418e0ec6cf
(31) https://docs.aws.amazon.com/general/latest/gr/iot-core.html
(34) https://docs.aws.amazon.com/streams/latest/dev/kpl-with-schemaregistry.html
(37) https://www.hivemq.com/blog/hivemq-cloud-vs-aws-iot/
(38) https://www.rabbitmq.com/protocols.html
(39) https://www.rabbitmq.com/jms-client.html
(40) https://activemq.apache.org/protocols
(41) https://activemq.apache.org/components/artemis/documentation/1.0.0/using-jms.html
(42) https://docs.aws.amazon.com/general/latest/gr/iot-core.html#message-broker-limits
(43) https://docs.aws.amazon.com/iot/latest/developerguide/binary-payloads.html
(44) https://docs.aws.amazon.com/kinesis/latest/APIReference/API_Record.html https://docs.aws.amazon.com/streams/latest/dev/fundamental-stream.html
(45) https://docs.aws.amazon.com/sns/latest/api/API_Publish.html
(46) https://docs.aws.amazon.com/AWSSimpleQueueService/latest/APIReference/API_SendMessage.html
(47) https://docs.aws.amazon.com/eventbridge/latest/APIReference/API_PutEvents.html
(48) https://kafka.apache.org/protocol#protocol_types
(49) https://redis.io/commands/XADD; https://redis.io/topics/data-types-intro
(50) https://docs.aws.amazon.com/streams/latest/dev/kinesis-low-latency.html
(51) https://www.pubnub.com/blog/amazon-sns-pubnub-differences-pubsub/
(52) https://www.confluent.io/blog/kafka-fastest-messaging-system/
(53) https://softwaremill.com/mqperf/
(54) https://www.rabbitmq.com/reliability.html
(55) https://kafka.apache.org/0110/documentation/streams/core-concepts
(56) https://medium.com/@andy.bryant/processing-guarantees-in-kafka-12dd2e30be0e
(57) https://activemq.apache.org/
(58) https://activemq.apache.org/performance.html
(59) https://activemq.apache.org/configuring-wire-formats.html see maxFrameSize = MAX_LONG - theoretical maximum
(61) https://www.cloudamqp.com/blog/what-is-the-message-size-limit-in-rabbitmq.html
(62) https://docs.aws.amazon.com/sns/latest/dg/fifo-message-filtering.html
(64) https://www.amazonaws.cn/en/kinesis/data-streams/features/
(65) https://aws.amazon.com/iot-core/faqs/
(66) https://www.cloudamqp.com/blog/when-to-use-rabbitmq-or-apache-kafka.html
(67) https://www.cloudamqp.com/blog/part4-rabbitmq-for-beginners-exchanges-routing-keys-bindings.html
(68) https://aws.amazon.com/blogs/compute/integrating-amazon-mq-with-other-aws-services-via-apache-camel/
(69) https://camel.apache.org/manual/latest/faq/how-does-camel-work-with-activemq.html https://camel.apache.org/components/3.8.x/activemq-component.html
(70) https://redis.io/topics/pubsub
(72) https://stackoverflow.com/questions/60019790/redis-streams-with-persistent-storage
(73) https://redislabs.com/ebook/part-2-core-concepts/chapter-3-commands-in-redis/3-6-publishsubscribe/
(74) https://redis.io/commands/xread
(75) https://redis.io/topics/clients - for the given case the max message size needs to be calculate backwards from the 32Mb/min limit.
(76) Assuming an argument of XADD is a string https://redis.io/commands/xadd, infer a max string size from https://redis.io/topics/data-types
(77) https://docs.amazonaws.cn/en_us/streams/latest/dev/building-consumers.html