Details
-
Improvement
-
Resolution: Done
-
Major
-
Morpheus, 7.6.2
-
0
Description
We finished running XDCR/SGW tests with 10M and 100M docs. Here is a table with the results we got so far and our observations:
Test Case | Replication Rate (docs replicated/sec) 10M docs (7.6.2-3674) |
Replication Rate (docs replicated/sec) 10M docs (8.0.0-1649) |
Replication Rate (docs replicated/sec) 100M docs (7.6.2-3674) |
Replication Rate (docs replicated/sec) 100M docs (8.0.0-1649) |
Observations |
---|---|---|---|---|---|
Baseline (SGW not connected to cluster) | 93,978 | 101,807 | 102,312 | 105,223 | Baseline |
SGWs connected to clusters, but SGW import disabled | 98,728 | 97,731 | 99,463 | 104,748 | Different to previous case, as SGW docs are tracked and filtered out |
SGWs connected to clusters, SGW import enabled for (SGW connected to) source cluster | 82,915 | 88,160 | 85,536 | 87,141 | Regression seems to be caused by increased number of mutations to process (as every imported doc is another mutations) |
SGWs connected to clusters, SGW import enabled for (SGW connected to) target cluster | 58,623 | 68,478 | 58,335 | 104,139 | Regression seems to be caused by extra CPU usage due to SGW import |
SGWs connected to clusters, SGW import enabled for (SGWs connected to) both source and target cluster | 52,903 | 69,422 | 84,557 | 87,713 | Regressions from the previous 2 test cases impacting replication at the same time |
The numbers are not final, as there is still work to be done for XDCR/SGW Coexistence, but I think it's worth having a ticket to make looking for results easier.