Performance Benchmarking with TPC-C

This page shows you how to reproduce . Across all scales, CockroachDB can process tpmC (new order transactions per minute) at near maximum efficiency. Start by choosing the scale you’re interested in:

Workload	Cluster size	Warehouses	Data size
Local	3 nodes on your laptop	10	2 GB
Local (multi-region)	9 in-memory nodes on your laptop using cockroach demo	10	2 GB
Small	3 nodes on c5d.4xlarge machines	2500	200 GB
Medium	15 nodes on c5d.4xlarge machines	13,000	1.04 TB
Large	81 nodes on c5d.9xlarge machines	140,000	11.2 TB

Before you begin

• TPC-C provides the most realistic and objective measure for OLTP performance at various scale factors. Before you get started, consider reviewing .
• Make sure you have already .

Step 1. Start CockroachDB

The --insecure flag used in this tutorial is intended for non-production testing only. To run CockroachDB in production, use a secure cluster instead.

1. In separate terminal windows, use the command to start 3 nodes:

   $ cockroach start \
   --insecure \
   --store=tpcc-local1 \
   --listen-addr=localhost:26257 \
   --http-addr=localhost:8080 \
   --join=localhost:26257,localhost:26258,localhost:26259
   
   $ cockroach start \
   --insecure \
   --store=tpcc-local2 \
   --listen-addr=localhost:26258 \
   --http-addr=localhost:8081 \
   --join=localhost:26257,localhost:26258,localhost:26259
   
   $ cockroach start \
   --insecure \
   --store=tpcc-local3 \
   --listen-addr=localhost:26259 \
   --http-addr=localhost:8082 \
   --join=localhost:26257,localhost:26258,localhost:26259
   

2. Use the command to perform a one-time initialization of the cluster:

   $ cockroach init \
   --insecure \
   --host=localhost:26257
   

Step 2. Import the TPC-C dataset

CockroachDB comes with a number of for simulating client traffic. This step features CockroachDB’s version of the TPC-C workload.

Use to load the initial schema and data:

$ cockroach workload fixtures import tpcc \
--warehouses=10 \
'postgresql://root@localhost:26257?sslmode=disable'

This will load 2 GB of data for 10 “warehouses”.

Step 3. Run the benchmark

Run the workload for ten “warehouses” of data for ten minutes:

$ cockroach workload run tpcc \
--warehouses=10 \
--ramp=3m \
--duration=10m \
'postgresql://root@localhost:26257?sslmode=disable'

You’ll see per-operation statistics every second:

Initializing 20 connections...
Initializing 100 workers and preparing statements...
_elapsed___errors__ops/sec(inst)___ops/sec(cum)__p50(ms)__p95(ms)__p99(ms)_pMax(ms)
    1.0s        0            0.0            0.0      0.0      0.0      0.0      0.0 delivery
    1.0s        0            0.0            0.0      0.0      0.0      0.0      0.0 newOrder
...
  105.0s        0            0.0            0.2      0.0      0.0      0.0      0.0 delivery
  105.0s        0            4.0            1.8     44.0     46.1     46.1     46.1 newOrder
  105.0s        0            0.0            0.2      0.0      0.0      0.0      0.0 orderStatus
  105.0s        0            1.0            2.0     14.7     14.7     14.7     14.7 payment
  105.0s        0            0.0            0.2      0.0      0.0      0.0      0.0 stockLevel
...

For more tpcc options, use cockroach workload run tpcc --help. For details about other built-in load generators, use cockroach workload run --help.

Step 4. Interpret the results

Once the workload has finished running, you’ll see a final output line:

_elapsed_______tpmC____efc__avg(ms)__p50(ms)__p90(ms)__p95(ms)__p99(ms)_pMax(ms)
  300.0s      121.6  94.6%     41.0     39.8     54.5     71.3     96.5    130.0

You will also see some audit checks and latency statistics for each individual query. For this run, some of those checks might indicate that they were SKIPPED due to insufficient data. For a more comprehensive test, run workload for a longer duration (e.g., two hours). The tpmC (new order transactions/minute) number is the headline number and efc (“efficiency”) tells you how close CockroachDB gets to theoretical maximum tpmC.

The TPC-C specification has p90 latency requirements in the order of seconds, but as you see here, CockroachDB far surpasses that requirement with p90 latencies in the tens of milliseconds.

Step 5. Clean up

1. When you’re done with your test cluster, stop the nodes.

   Get the process IDs of the nodes:

   ps -ef | grep cockroach | grep -v grep
   
     501  4482     1   0  2:41PM ttys000    0:09.78 cockroach start --insecure --store=tpcc-local1 --listen-addr=localhost:26257 --http-addr=localhost:8080 --join=localhost:26257,localhost:26258,localhost:26259
     501  4497     1   0  2:41PM ttys000    0:08.54 cockroach start --insecure --store=tpcc-local2 --listen-addr=localhost:26258 --http-addr=localhost:8081 --join=localhost:26257,localhost:26258,localhost:26259
     501  4503     1   0  2:41PM ttys000    0:08.54 cockroach start --insecure --store=tpcc-local3 --listen-addr=localhost:26259 --http-addr=localhost:8082 --join=localhost:26257,localhost:26258,localhost:26259
   

   Gracefully shut down each node, specifying its process ID:

   kill -TERM 4482
   
   kill -TERM 4497
   

For the last node, the shutdown process will take longer (about a minute) and will eventually stop the node. This is because, with only 1 of 3 nodes left, all ranges no longer have a majority of replicas available, and so the cluster is no longer operational.

kill -TERM 4503

2. To restart the cluster at a later time, run the same cockroach start commands as earlier from the directory containing the nodes’ data stores.

   If you do not plan to restart the cluster, you may want to remove the nodes’ data stores:

   $ rm -rf tpcc-local1 tpcc-local2 tpcc-local3