How can Kafka Streams throughput be increased?

Question

Currently ~70 records/s will be processed on a single node with a single Kafka broker. The throughput is low, as is the CPU utilisation and the memory usage. My topology:

projekte
    .leftJoin(wirtschaftseinheiten)
    .leftJoin(mietobjekte)
    .cogroup { _, current, previous: ProjektAggregat ->
        previous.copy(
            projekt = current.projekt,
            wirtschaftseinheit = current.wirtschaftseinheit,
            mietobjekt = current.mietobjekt,
            projektErstelltAm = current.projektErstelltAm
        )
    }
    .cogroup(projektstatus.groupByKey()) { _, projektstatusEvent, aggregat -> aggregat + projektstatusEvent }
    .cogroup(befunde.groupByKey()) { _, befundAggregat, aggregat -> aggregat + befundAggregat }
    .cogroup(aufgaben.groupByKey()) { _, aufgabeAggregat, aggregat -> aggregat + aufgabeAggregat }
    .cogroup(durchfuehrungen.groupByKey()) { _, durchfuehrungAggregat, aggregat -> aggregat + durchfuehrungAggregat }
    .cogroup(gruppen.groupByKey()) { _, gruppeAggregat, aggregat -> aggregat + gruppeAggregat }
    .aggregate({ ProjektAggregat() }, Materialized.`as`(projektStoreSupplier))

I've tried to increase different size to feed more data to my stream:

• cache.max.bytes.buffering: 52428800
• max.request.size: 52428800 but they didn't measurably help.

How can I increase throughput to achieve optimal system utilisation?

Answer 1

With the following properties I could speed up processing by factor 4 by increasing I/O throughput:

# consumer
max.partition.fetch.bytes: 52428800
max.poll.records: 5000
fetch.min.bytes: 5242880
#producer
max.request.size: 52428800
batch.size: 200000
linger.ms: 100
# common
cache.max.bytes.buffering: 52428800

Increasing parallelism to 20 on a 24-core system I further could achieve an optimal CPU utilisation of 80% and speed up the processing further by a factor of 9:

num.stream.threads: 20