What Factors Affect the DRS Task Speed and How Do I Estimate the Time Required?

Factors

• Read throughput of the source database

  The higher the read throughput is, the faster the migration speed can be, and the less the time required can be. Factors that affect the throughput include but are not limited to server specifications, load, disk I/O performance, and database traffic limiting.

• Write throughput of the destination database

  The higher the write throughput is, the faster the migration speed can be, and the less the time required can be. Factors that affect the throughput include but are not limited to server specifications, load, disk I/O performance, and database traffic limiting.

• Available network throughput

  The higher the available network throughput is, the faster the migration speed can be, and the less the time required can be. Factors that affect network throughput include but are not limited to available bandwidth, firewalls, and network device traffic limiting.

• Network quality and delay

  The shorter the network latency is, the faster the migration speed can be, and the less the time required can be. The factors include but are not limited to the distance between the source or destination database and the DRS instance. Poor network quality (for example, high packet loss rate) reduces the migration speed.

• DRS instance specifications

  The larger the DRS instance specifications are, the faster the migration speed can be, and the less the time required can be.

• Model and distribution of source data

  Such factors include whether there is a primary key, whether there is a partition table, whether there is a heap table, average data volume in a single row, number of tables, and number of indexes.

• Whether there is data in the destination database

  Existing data in the destination database may cause data conflicts during migration, resulting in performance deterioration.

• Whether the destination database has a trigger

  If the destination database has a trigger, the write performance may deteriorate during migration.

• Destination database backup and log settings

  If not necessary, disable destination database backup and transaction logs during the migration to improve migration performance.

• Incremental data generation speed of the source database

  The faster the incremental data is generated in the source database, the longer it takes to balance the incremental data.

• Number of DRS tasks

  If performance bottlenecks caused by other factors are not considered, you can split DRS tasks by table to improve the overall migration performance.

Estimated Migration Duration

There are many factors that affect the migration duration. No common calculation method can be used to evaluate the migration duration. You can evaluate the migration duration based on the following data. You are advised to create a test task in an environment with the same specifications, load, network configuration, and data model as the instance to be migrated to evaluate the migration duration.

DRS migration speed in the full migration phase:

• MySQL serving as the source: 50 MB/s
• Oracle serving as the source: 40 MB/s
• Redis serving as the source in the full+incremental scenario: 30 MB/s
• GaussDB serving as the source: 40 MB/s
• DDM serving as the source: 20 MB/s
• MongoDB serving as the source: 20 MB/s
• PostgreSQL serving as the source: 30 MB/s

• There are many factors that affect the DRS migration speed. The current migration speed is the test data when there is no network and database performance bottlenecks and the task specifications are large. The migration speed is for reference only.
• When the destination database is Oracle or GaussDB(DWS), the migration speed in the full phase is 30% to 50% lower than that of other types of databases due to the write mechanism of the destination database.
• The write performance of the MongoDB database is affected by the number of indexes. The write performance decreases by 5% to 8% when there is one index. The more indexes, the slower the write speed.