1.30 0.58 600 2.11 0.96 900 3.07 1.37 1200 4.10 1.84 1500 5.18 2.30 Sysbench - Update Non-index (TPS) Thread TiDB 3.0 Amazon Aurora 150 24,710.55 9341.03 300 30,347.88 12,052.74 600 38,685.87 11,834.75 567.08 11,855.36 1200 49,982.75 12,065.85 1500 55,717.09 12,100.72 Sysbench - Update Non-index (TPS) Sysbench - Update Non-index (.95 Latency) Thread TiDB 3.0 Amazon Aurora 150 8.43 20.00 300 13.95 25 600 23.52 71.83 900 30.81 104.84 1200 39.65 137.35 1500 51.02 155.80 Sysbench - Update Index (TPS) Thread TiDB 3.0 Amazon Aurora 150 15,202.94 8953.53 300 18,874.35 10,326.72 600 23,882.45 10,164.60

data exceeding affordable threshold. What’s more, database sharding can also effectively disperse TPS. Table sharding, though cannot ease the database pressure, can provide possibilities to transfer distributed Splitting data through database sharding and table sharding is an effective method to deal with high TPS and mass amount data system, because it can keep the data amount lower than the threshold and evacuate Readwrite-splitting 8.3.1 Background Database throughput has faced the bottleneck with increasing TPS. For the application with massive concurrence read but less write in the same time, we can divide

data exceeding affordable threshold. What’s more, database sharding can also effectively disperse TPS. Table sharding, though cannot ease the database pressure, can provide possibilities to transfer distributed Splitting data through database sharding and table sharding is an effective method to deal with high TPS and mass amount data system, because it can keep the data amount lower than the threshold and evacuate Readwrite-splitting 3.3.1 Background Database throughput has faced the bottleneck with increasing TPS. For the application with massive concurrence read but less write in the same time, we can divide

data exceeding affordable threshold. What’s more, database sharding can also effectively disperse TPS. Table sharding, though cannot ease the database pressure, can provide possibilities to transfer distributed Splitting data through database sharding and table sharding is an effective method to deal with high TPS and mass amount data system, because it can keep the data amount lower than the threshold and evacuate Readwrite-splitting 3.3.1 Background Database throughput has faced the bottleneck with increasing TPS. For the application with massive concurrence read but less write in the same time, we can divide

VARBINARY(10000) NOT NULL, pos VARBINARY(10000) NOT NULL, stop_pos VARBINARY(10000) DEFAULT NULL, max_tps BIGINT(20) NOT NULL, max_replication_lag BIGINT(20) NOT NULL, cell VARBINARY(1000) DEFAULT NULL, tablet_types sh', 'VReplicationExec', 'test -200', """insert into _vt.vreplication (db_name , source , pos, max_tps , max_replication_lag , tablet_types , time_updated , transaction_timestamp , state) values ('vt_keyspace' typical position would look like this MySQL56/ac6c45eb-71c2-11e9-92ea-0a580a1c1026:1-1296. • max_tps: 99999, reserved. • max_replication_lag: 99999, reserved. • tablet_types: specifies a comma separated

VARBINARY(10000) NOT NULL, pos VARBINARY(10000) NOT NULL, stop_pos VARBINARY(10000) DEFAULT NULL, max_tps BIGINT(20) NOT NULL, max_replication_lag BIGINT(20) NOT NULL, cell VARBINARY(1000) DEFAULT NULL, tablet_types 'VReplicationExec', 'test -200', """insert into _vt.vreplication 128 (db_name , source , pos, max_tps , max_replication_lag , tablet_types , time_updated , transaction_timestamp , state) values ('vt_keyspace' typical position would look like this MySQL56/ac6c45eb-71c2-11e9-92ea-0a580a1c1026:1-1296. • max_tps: 99999, reserved. • max_replication_lag: 99999, reserved. • tablet_types: specifies a comma separated

session Before: insert test data finished ... elapse : 34.563343s tps:28932.386438 After: insert test data finished ... elapse : 28.859153s tps:34651.051825 sync.Pool • Thread safe • Reuse objects to relieve

流程体系 • 管理：服务集成，统一管理 应用互联互通 应用形态复杂 • KPI: 峰值CPU利用率不低 于30% • 资源申请：按峰值30%进 行申请 • 峰值：1000TPS, 平时： 100TPS • 做自己擅长的事情，合作 方式开发 • 产品迭代：如何持续演进 和优化 • 外包管理：如何标准化降 低管理成本，提高质量 外包开发模式 资源利用率KPI 01 04

VARBINARY(10000) NOT NULL, pos VARBINARY(10000) NOT NULL, stop_pos VARBINARY(10000) DEFAULT NULL, max_tps BIGINT(20) NOT NULL, max_replication_lag BIGINT(20) NOT NULL, cell VARBINARY(1000) DEFAULT NULL, tablet_types 'VReplicationExec', 'test -200', """insert into _vt.vreplication 293 (db_name , source , pos, max_tps , max_replication_lag , tablet_types , time_updated , transaction_timestamp , state) values ('vt_keyspace' typical position would look like this MySQL56/ac6c45eb-71c2-11e9-92ea-0a580a1c1026:1-1296. • max_tps: 99999, reserved. • max_replication_lag: 99999, reserved. • tablet_types: specifies a comma separated

VARBINARY(10000) NOT NULL, pos VARBINARY(10000) NOT NULL, stop_pos VARBINARY(10000) DEFAULT NULL, max_tps BIGINT(20) NOT NULL, max_replication_lag BIGINT(20) NOT NULL, cell VARBINARY(1000) DEFAULT NULL, tablet_types 'localhost:15999', 'VReplicationExec', 'test -200', """insert into _vt.vreplication (db_name , source , pos, max_tps , max_replication_lag , tablet_types , time_updated , transaction_timestamp , state) values ('vt_keyspace' typical position would look like this MySQL56/ac6c45eb-71c2-11e9-92ea-0a580a1c1026:1-1296. • max_tps: 99999, reserved. • max_replication_lag: 99999, reserved. • tablet_types: specifies a comma separated