I am using NHibernate 1.2.0 GA.

This is less of a question and more my view on performance using Generic Collections in NHibernate.

I am running performance tests at the moment.

I have an table of objects in the database called CARs for instance.

If I do a criteria query on them using generics it takes 68 seconds to retrieve 100,000 rows.

criteria.List<Car>();

If I simply remove the Generics and do this.

criterial.List();

It now takes 42 seconds. The object coming back as around 42 columns so its big enough. The Query itself is very fast, it is doing an outer joing on one tabe that has a one-to-zero-or-one mapping. Its the marshalling that takes most of the time. Logging is OFF by the way.

This is a dramatic decrese in time and now it is actually nearly as fast as ADO.NET direct. I removed the generic collections for ADO.NET (This is mapping code that I wrote myself) and there was no performance increase.

This leads me to believe that whatever way NHibernate is using generics is not great or needs to be optimized further.

Anyone got anything to add to this. Are there further ways of improving performance.

Maybe you can use a profiler. What do you think?

In this case I would be performing profiling on NHibernate code. I cant optimize that.

I need ideas and ways to improve how NHibernate Marshalls objects from the database.

IQuery : Has roughly the same performance.
Enumerable : Has much worse performance.

I could use the second level cache and the query cache but I am just interested in the initial hit on the database and getting that data marshalled.

Once I am happy with its performance I will then look at the second level chache.

You can, if it's not optimized yet (I believe it is) and if you know how to. But even if you can't or don't want to, if you find something relevant NHibernate team could do this in the future.

Looking at the source code... the implementations of List() and List<T>() are identical other than one constructs an ArrayList and the other a List<T>. Both methods delegate to the same method which takes an IList to fill. Maybe there's performance hit to adding items to a List<T> via its IList interface....

Actually. For some reason, on changing it back to List<T>() the performance gain I had earlier is not going away.

Seems like maybe this performance gain may be down to something else. I am going to investigate further.