In one of the sites, I came across a performance issue where data loading task was taking more than usual. After analysing the AWR report, we observed that network wait was high and the task was waiting for the event sql*net more data from client.
As you can see, in below AWR there are two consecutive days screenshot, and in both the AWR there was backlogs. But in second AWR where average wait in ms is 158, processing speed was little bit faster than first one where average wait in ms was 293.
As observed in AWR, network wait is because of the the shadow process has received part of a call from the client process (for example, SQL*Plus, Pro*C, and JDBC) in the first network package and is waiting for more data for the call to be complete. Examples are large SQL or PL/SQL block and insert statements with large amounts of data.
The possible cause might be network latency problems, tcp_no_delay configuration issues and large array insert.