Sign InTry Free

TiDB Sysbench Performance Test Report -- v6.2.0 vs. v6.1.0

Test overview

This test aims at comparing the Sysbench performance of TiDB v6.2.0 and TiDB v6.1.0 in the Online Transactional Processing (OLTP) scenario. The results show that performance of v6.2.0 is basically the same as that of v6.1.0. The performance of Point Select slightly drops by 3.58%.

Test environment (AWS EC2)

Hardware configuration

Service typeEC2 typeInstance count

Software version

Service typeSoftware version
PDv6.1.0 and v6.2.0
TiDBv6.1.0 and v6.2.0
TiKVv6.1.0 and v6.2.0

Parameter configuration

TiDB v6.2.0 and TiDB v6.1.0 use the same configuration.

TiDB parameter configuration

log.level: "error" prepared-plan-cache.enabled: true tikv-client.max-batch-wait-time: 2000000

TiKV parameter configuration

storage.scheduler-worker-pool-size: 5 3 raftstore.apply-pool-size: 3 rocksdb.max-background-jobs: 8 server.grpc-concurrency: 6 readpool.unified.max-thread-count: 10

TiDB global variable configuration

set global tidb_hashagg_final_concurrency=1; set global tidb_hashagg_partial_concurrency=1; set global tidb_enable_async_commit = 1; set global tidb_enable_1pc = 1; set global tidb_guarantee_linearizability = 0; set global tidb_enable_clustered_index = 1; set global tidb_prepared_plan_cache_size=1000;

HAProxy configuration - haproxy.cfg

For more details about how to use HAProxy on TiDB, see Best Practices for Using HAProxy in TiDB.

global # Global configuration. pidfile /var/run/ # Writes the PIDs of HAProxy processes into this file. maxconn 4000 # The maximum number of concurrent connections for a single HAProxy process. user haproxy # The same with the UID parameter. group haproxy # The same with the GID parameter. A dedicated user group is recommended. nbproc 64 # The number of processes created when going daemon. When starting multiple processes to forward requests, ensure that the value is large enough so that HAProxy does not block processes. daemon # Makes the process fork into background. It is equivalent to the command line "-D" argument. It can be disabled by the command line "-db" argument. defaults # Default configuration. log global # Inherits the settings of the global configuration. retries 2 # The maximum number of retries to connect to an upstream server. If the number of connection attempts exceeds the value, the backend server is considered unavailable. timeout connect 2s # The maximum time to wait for a connection attempt to a backend server to succeed. It should be set to a shorter time if the server is located on the same LAN as HAProxy. timeout client 30000s # The maximum inactivity time on the client side. timeout server 30000s # The maximum inactivity time on the server side. listen tidb-cluster # Database load balancing. bind # The Floating IP address and listening port. mode tcp # HAProxy uses layer 4, the transport layer. balance leastconn # The server with the fewest connections receives the connection. "leastconn" is recommended where long sessions are expected, such as LDAP, SQL and TSE, rather than protocols using short sessions, such as HTTP. The algorithm is dynamic, which means that server weights might be adjusted on the fly for slow starts for instance. server tidb-1 check inter 2000 rise 2 fall 3 # Detects port 4000 at a frequency of once every 2000 milliseconds. If it is detected as successful twice, the server is considered available; if it is detected as failed three times, the server is considered unavailable. server tidb-2 check inter 2000 rise 2 fall 3 server tidb-3 check inter 2000 rise 2 fall 3

Test plan

  1. Deploy TiDB v6.2.0 and v6.1.0 using TiUP.
  2. Use Sysbench to import 16 tables, each table with 10 million rows of data.
  3. Execute the analyze table statement on each table.
  4. Back up the data used for restore before different concurrency tests, which ensures data consistency for each test.
  5. Start the Sysbench client to perform the point_select, read_write, update_index, and update_non_index tests. Perform stress tests on TiDB via HAProxy. For each concurrency under each workload, the test takes 20 minutes.
  6. After each type of test is completed, stop the cluster, overwrite the cluster with the backup data in step 4, and restart the cluster.

Prepare test data

Run the following command to prepare the test data:

sysbench oltp_common \ --threads=16 \ --rand-type=uniform \ --db-driver=mysql \ --mysql-db=sbtest \ --mysql-host=$aws_nlb_host \ --mysql-port=$aws_nlb_port \ --mysql-user=root \ --mysql-password=password \ prepare --tables=16 --table-size=10000000

Perform the test

Run the following command to perform the test:

sysbench $testname \ --threads=$threads \ --time=1200 \ --report-interval=1 \ --rand-type=uniform \ --db-driver=mysql \ --mysql-db=sbtest \ --mysql-host=$aws_nlb_host \ --mysql-port=$aws_nlb_port \ run --tables=16 --table-size=10000000

Test results

Point Select performance

Threadsv6.1.0 TPSv6.2.0 TPSv6.1.0 95% latency (ms)v6.2.0 95% latency (ms)TPS improvement (%)

Compared with v6.1.0, the Point Select performance of v6.2.0 slightly drops by 3.58%.

Point Select

Update Non-index performance

Threadsv6.1.0 TPSv6.2.0 TPSv6.1.0 95% latency (ms)v6.2.0 95% latency (ms)TPS improvement (%)

Compared with v6.1.0, the Update Non-index performance of v6.2.0 is basically unchanged, reduced by 0.85%.

Update Non-index

Update Index performance

Threadsv6.1.0 TPSv6.2.0 TPSv6.1.0 95% latency (ms)v6.2.0 95% latency (ms)TPS improvement (%)

Compared with v6.1.0, the Update Index performance of v6.2.0 is basically unchanged, reduced by 0.47%.

Update Index

Read Write performance

Threadsv6.1.0 TPSv6.2.0 TPSv6.1.0 95% latency (ms)v6.2.0 95% latency (ms)TPS improvement (%)

Compared with v6.1.0, the Read Write performance of v6.2.0 is reduced by 1.21%.

Read Write

Download PDF
One-stop & interactive experience of TiDB's capabilities WITHOUT registration.
TiDB Dedicated
TiDB Serverless
Get Demo
Get Started
© 2024 PingCAP. All Rights Reserved.
Privacy Policy.