I'm comparing Sql Server and Oracle performance of reads using C#. Basically, I'm using ADO.NET and ODP.NET and obtaining similar values on both, except when I read BLOB columns on distributed machines (one machine requesting another with DB).
Important notes:
- The results are even for Sql and Oracle when testing locally.
- Latency between machines are '<1ms'
In my test, all the data are exactly the same, and I'm reading 2881 rows with some binary data on each row. The obtained results are:
- Sql Server locally: 1.6s
- Oracle locally: 1.8s (good result)
- Sql Server distributed: 1.8s
- Oracle distributed: 10.0s (problem here!)
Configurations that I'm already using on Oracle to perform better:
OracleCommand.InitialLOBFetchSize = -1;(tried other positive values but with worst results)OracleCommand.InitialLONGFetchSize = -1;(tried other positive values but with worst results)OracleReader.FetchSize = OracleReader.RowSize * 10000;(setting before the firstRead())
So the question is:
- There is any other configuration I'm missing to obtain better performance when reading BLOB columns?
Other notes:
- I assumed
InitialLOBFetchSize = -1would improve this scenario considering low latency, but didn't improved too much (withInitialLOBFetchSize = 0time was ~12.8s). - Already tried some network improvement configurations on Oracle instance, but without expressive results
- Using Oracle.DataAccess.dll x64 v4.121.2.0
- Tried using managed and unmanaged Oracle drivers, with better results with the unmanaged ones (the one I was already using since the beginning)
References:
