When running BMF against a large database, there are deadlocks. Is there is an optimal proportion of virtual users/db size to use to avoid deadlocking?
What type of test are you running? For some of the standard benchmarks, there is a relation between scale factor (db size) and number of virtual users. For example, tpc-c has a 1:10 ratio between scale factor and virtual users.
If they are running the TPC-C benchmark there is a recommended limit of 10 virtual users per scale factor. Now deadlocks may still happen but they are much more likely if outside this recommendation. Also the default latencies of the TPC-C benchmark need to be preserved to avoid an increase likelihood of deadlocks.
The scale factor for the Scalable Hardware test not only specifies the dataset size but also for how many users that there queries will return data. In other words for a scale factor of 3 only the first 3 virtual users will return/update rows for thier queries. All the other virtual users (4 and above) will have thier queries execute but not return/update any data.