Lost in translation
Even when systems speak the same standard protocol, there might be a whole bunch of translations and copies happening under the hood. ODBC, for all its benefits, was still designed during a time when it was much more common to be requesting wide tables with large numbers of columns and fewer rows as compared to modern data analysis. While it enabled connectivity between different disparate systems, there's still a lot of translating and copying that has to happen in the ODBC drivers for everything to work correctly. Figure 3.1 shows a comparison between a standard data science workflow using typical ODBC or JDBC and using the Arrow-JDBC adapter.
cLook first at the left side of Figure 3.1, the typical case when using JDBC. There are three points where data has to be translated between formats, as follows:
- First, data is translated inside the JDBC/ODBC driver from whatever format the database speaks natively into the JDBC/ODBC standards...
