I have an ADF pipeline that copies tables from an on-prmemise SQL Server to Azure SQL Db. The problem is that throughput when running the copy activity will always be ~5MB/s. I tried using increased DIU, partitioning, parallelism and increasing VM size to improve the throughput. This did decrease writing to sink time (to azure sql) but not reading from source (on-premise sql). An azure integration runtime is used for this setup, together with an azure VM for network forwarding, as described here. The on premise database is able to return data much faster in SSMS and if querying from an azure function.