Performance Benchmarks

This document summarizes the sync performance benchmarks of Apache Doris CCR (Cross Cluster Replication) under the default configuration. The tests cover three typical scenarios: full sync, incremental sync, and Flink real-time write sync. The results can serve as a reference for capacity planning and SLA evaluation.

If you face scenarios such as high network latency or high-throughput writes, see the Operations Manual for parameter tuning.

Applicable Scenarios

Scenario	Metrics of Interest	Corresponding Section
Evaluate whether CCR meets the business SLA	Sync latency, first-time sync duration	Full Sync Performance Test, Incremental Sync Performance Test
Evaluate real-time data replication capability	Real-time write lag	Flink Sync Performance Test
Capacity planning for cross-cluster migration	Full initialization duration	Full Sync Performance Test
Baseline reference before tuning	Throughput and latency under default configuration	All test conclusions

Test Environment

Test Dataset

• Dataset: TPC-H 1T

Test Cluster Configuration

The upstream and downstream clusters use identical specifications:

Item	Upstream Configuration	Downstream Configuration
FE	2 cores, 16 GB	2 cores, 16 GB
BE	3 nodes, 16 cores, 64 GB, 3*500 GB per node	3 nodes, 16 cores, 64 GB, 3*500 GB per node

---

Incremental Sync Performance Test

Test goal: Measure the time gap between an upstream write completing and the data becoming visible downstream after a sync relationship is established.

Test Steps

1. Create the TPC-H 1T database and table definitions on the upstream cluster.
2. Create a sync task for the TPC-H 1T database.
3. Wait for the TPC-H 1T data import to complete, and record the completion time.
4. Wait for the downstream sync to complete, and record the completion time.

Test Conclusion

Incremental sync time gap: 33 seconds.

---

Full Sync Performance Test

Test goal: Measure the total time from creating a new sync task to the downstream data becoming fully visible, given that the upstream already holds the complete 1T dataset.

Test Steps

1. Create the TPC-H 1T database and table definitions on the upstream cluster, complete the data import, and record the completion time.
2. Create a sync task for the TPC-H 1T database.
3. Wait for the downstream sync to complete, and record the completion time.

Test Conclusion

Full sync time gap: 6 minutes 1 second.

---

Flink Sync Performance Test

Test goal: Measure the real-time gap (lag) between downstream sync progress and upstream write progress when the upstream is continuously writing through Flink.

Test Steps

1. The upstream imports 100,000,000 rows using Flink.
2. Create the sync task for the database and tables.
3. At each stage, observe the difference between the downstream sync completion time and the upstream import completion time (for example: 1,000,000 rows, 2,000,000 rows, and so on).
4. Record the completion time of the last upstream import.
5. Record the downstream sync completion time.

Test Conclusion

The lag time at each stage stays within 5 seconds.

---

Summary of Test Conclusions

Test Scenario	Data Scale	Sync Latency / Duration
Incremental sync	TPC-H 1T	33 seconds
Full sync	TPC-H 1T	6 minutes 1 second
Flink real-time sync	100 million rows streamed	Lag at each stage ≤ 5 seconds

The numbers above are measured under the default cluster configuration listed on this page. In real production environments, sync performance depends on cluster specifications, network bandwidth, the number of tablets, and the business write pattern. For further optimization, see the Operations Manual.