Provide commentary on the latency improvements in v4.3 for this test run.
- Cosmos account configured with session consistency and in a single region - West US
- Client running on a local desktop in San Diego, CA
- Max provisioned RUs = 400
- MaxRetryWaitTimeInSeconds = 10, MaxRetryAttempts = 100 and MaxPoolSize = 10
- Test executed for the following payload sizes: 1000, 5000, 10000, 50000, 100000, 200000 and 400000 bytes
- Each replace and upsert test adds an additional prefix called 'replaced' and 'upserted'
- RUs are collected using Histograms and that can only be an int or a long. Hence, I am using the Math.round() function to round off the RUs
- Test run executed with operation = SQL_ALL
- Documents created with a single partition key index.
- The following observations are based on the session consistency level.
- Data is available for strong, bounded staleness and eventual consistency levels if needed.
-
End to end latency numbers have improved substantially between v4.1 vs v4.3. This is applicable when upgrading from v4.2 as well.
-
Intermittent spikes in latency times have been eliminated
-
Sync create operations
%20comparison%20between%20v4.1%20and%20v4.3.png)
- Async create operations
%20comparison%20between%20v4.1%20and%20v4.3.png)
- Sync upsert operations
%20comparison%20between%20v4.1%20and%20v4.3.png)
- Async upsert operations
%20comparison%20between%20v4.1%20and%20v4.3.png)
- Sync point read operation
%20comparison%20between%20v4.1%20and%20v4.3.png)
- Async point read operation
%20comparison%20between%20v4.1%20and%20v4.3.png)
- Sync partition key based lookup
%20comparison%20between%20v4.1%20vs%20v4.3.png)
- Async partition key based lookup
%20comparison%20between%20v4.1%20and%20v4.3.png)