Key-value workload

• YCQL

Use this benchmark to test the performance of YugabyteDB using a key-value workload.

Recommended configuration

Cluster configuration

For this benchmark, you will set up a three-node YugabyteDB cluster with a replication factor of 3.

Machine configuration

• Amazon Web Services (AWS)

  • Instance type: i3.4xlarge
  • Storage: 2 x 1.9 TB NVMe SSD (comes preconfigured with the instance)

• Google Cloud Platform (GCP)

  • Instance type: n1-standard-16
  • Storage: 2 x 375 GB SSD

• on-premises data center

  • Instance: 16 CPU cores
  • Storage: 1 x 200 GB SSD (minimum)
  • RAM size: 30 GB (minimum)

Benchmark tool

We will use the YugabyteDB Workload Generator to perform this benchmark.

Download the YugabyteDB workload generator JAR file (yb-sample-apps.jar) as follows:

wget https://github.com/yugabyte/yb-sample-apps/releases/download/v1.4.1/yb-sample-apps.jar -O yb-sample-apps.jar

To run the workload generator tool, you must have:

• Java runtime or JDK installed.
• Set the environment variable $ENDPOINTS to the IP addresses (including hosts and ports) for the nodes of the cluster.

ENDPOINTS="X.X.X.X:9042,X.X.X.X:9042,X.X.X.X:9042"

Run the write-heavy key-value workload

Run the key-value workload with higher number of write threads (representing write-heavy workload).

Load 1B keys of 256 bytes each across 256 writer threads

$ java -jar ./yb-sample-apps.jar  \
      --workload CassandraKeyValue   \
      --nodes $ENDPOINTS             \
      --nouuid                       \
      --value_size 256               \
      --num_threads_read 0           \
      --num_threads_write  256       \
      --num_unique_keys 1000000000

Expected results

Run the read-heavy key-value workload

Run the key-value workload with higher number of read threads (representing read-heavy workload).

Load 1M keys of 256 bytes and access them with 256 reader threads.

$ java -jar ./yb-sample-apps.jar  \
      --workload CassandraKeyValue   \
      --nodes $ENDPOINTS             \
      --nouuid                       \
      --value_size 256               \
      --num_threads_read 256         \
      --num_threads_write 0          \
      --num_unique_keys 1000000

Expected results

Batch write-heavy KV workload

Run the key-value workload in batch mode and higher number of write threads (representing batched, write-heavy workload).

Load 1B keys of 256 bytes each across 64 writer threads in batches of 25 each.

$ java -jar ./yb-sample-apps.jar      \
      --workload CassandraBatchKeyValue  \
      --nodes $ENDPOINTS                 \
      --nouuid                           \
      --batch_size 25                    \
      --value_size 256                   \
      --num_threads_read 0               \
      --num_threads_write 64             \
      --num_unique_keys 1000000000

Expected results