Spark streaming slow down when using large broadcast objects in UDF

Question

I do stream processing from Event Hub using Spark and faced the following problem. For each incoming message I need to do some calculations (stateless). The calculation algo