The Producer has buffers of records per topic partition which are sized at batch.size property. The smaller the batch size the less the throughput and if the batch size is too big, the memory will be wasted since that part of memory is allocated for batching. This is because the data will be sent before the batch size limit hits.

Using a larger batch.size makes compression more efficient and if a data is larger than the batch size, it will not be batched.

Under heavy load, data will most probably be batched. However under light load data may not be batched. In that case increasing linger.ms property can increase throughput by increasing batching with fewer requests and with an increased latency on producer send

The buffers are sent as fast as broker can keep up. And this can be limited by max.in.flight.requests.per.connection property and if this sets to 1, any subsequent send request will wait the previous one return result.

By default producer will wait all replicas to return result as the default value for acknowledge property is ack=all. By setting ack=1, only the broker that gets the request will send confirmation instead of waiting all in-sync replicas.

The producer property compression.type allows to set compression on producer level. Default value is none. This setting can set to none, gzip, snappy, or lz4. The compression is done by batch and improves with larger batch sizes.

End to end compression is also possible if the Kafka Broker config compression.type set to producer. Then the compressed data will be sent from a producer, then goes to the topic and returned to any consumer in the compressed format. This way compression only happens once and is reused by the broker and consumer. 

The Producer config property is request.timeout.ms default 30 seconds. It is a client property and causes the client to wait that much time for the server to respond to a request

The Producer config property retries causes to retry a request if producer does not get an ack from kafka broker. It defaults to 0.

The Producer config property partitioner.class sets the partition strategy. By default partitioner.class is set to org.apache.kafka.clients.producer.internals.DefaultPartitioner  

https://medium.com/inspiredbrilliance/kafka-basics-and-core-concepts-5fd7a68c3193