Performance Tuning for Client

AutoMQ does not rely on local disks; instead, it writes data directly to object storage. In comparison to writing to local disks, file creation operations in object storage exhibit higher latency. For example, our empirical data shows that when writing a 4MiB file to S3, the P99 latency is approximately 400 milliseconds. Due to this, in some throughput-prioritized scenarios, it is necessary to suitably adjust client parameters to meet the system's performance goals (Little’s Law).

Note:

The configurations listed in this document are for reference only. Users need to consider their specific scenario demands and system goals comprehensively to reasonably adjust the corresponding parameter values.

If your scenario is sensitive to message read and write latency, you can contact the AutoMQ team for enterprise support through this form. The AutoMQ enterprise edition supports single-digit millisecond low-latency data reads and writes.

Producer Configuration

Producer Configuration Option	Description	Recommended Value
linger.ms	The delay time for the Producer to batch send messages, enhancing the efficiency of each request by accumulating more messages.	100
metadata.max.age.ms	The forced refresh time for metadata to prevent routing errors due to metadata expiration.	60000
batch.size	The maximum number of bytes in a single batch, directly affecting the number of network requests and throughput.	1048576
max.request.size	The maximum number of bytes in a single request, limiting the size of messages the Producer can send.	16777216

Consumer Configuration

Consumer Configuration Options	Description	Recommended Value
metadata.max.age.ms	The forced refresh time for metadata to prevent route errors due to expired metadata.	60000
max.partition.fetch.bytes	Limits the maximum amount of data returned in a Fetch request from a single partition, working together with fetch.max.bytes to control fetch granularity.	8388608