Vote of Disconfidence to Entity Framework

A friend of mine has found the following problem with Entity Framework 4:

  • Two simple classes and one association between them (one to many):


  • One condition to filter out soft-deleted entities (WHERE Deleted = 0):


  • 100 records in the database;
  • A simple query:
   1: var l = ctx.Person.Include("Address").Where(x => (x.Address.Name == "317 Oak Blvd." && x.Address.Number == 926) || (x.Address.Name == "891 White Milton Drive" && x.Address.Number == 497) ...);

Will produce the following SQL:

   1: SELECT 
   2: [Extent1].[Id] AS [Id], 
   3: [Extent1].[FullName] AS [FullName], 
   4: [Extent1].[AddressId] AS [AddressId], 
   5: [Extent202].[Id] AS [Id1], 
   6: [Extent202].[Name] AS [Name], 
   7: [Extent202].[Number] AS [Number]
   8: FROM                                                                                                                                                                                                        [dbo].[Person] AS [Extent1]
   9: LEFT OUTER JOIN [dbo].[Address] AS [Extent2] ON ([Extent2].[Deleted] = 0) AND ([Extent1].[AddressId] = [Extent2].[Id])
  10: LEFT OUTER JOIN [dbo].[Address] AS [Extent3] ON ([Extent3].[Deleted] = 0) AND ([Extent1].[AddressId] = [Extent3].[Id])
  11: LEFT OUTER JOIN [dbo].[Address] AS [Extent4] ON ([Extent4].[Deleted] = 0) AND ([Extent1].[AddressId] = [Extent4].[Id])
  12: LEFT OUTER JOIN [dbo].[Address] AS [Extent5] ON ([Extent5].[Deleted] = 0) AND ([Extent1].[AddressId] = [Extent5].[Id])
  13: LEFT OUTER JOIN [dbo].[Address] AS [Extent6] ON ([Extent6].[Deleted] = 0) AND ([Extent1].[AddressId] = [Extent6].[Id])
  14: ...
  15: WHERE ((N'317 Oak Blvd.' = [Extent2].[Name]) AND (926 = [Extent3].[Number])) 
  16: ...

And will result in 680 MB of memory being taken!

Now, Entity Framework has been historically known for producing less than optimal SQL, but 680 MB for 100 entities?!

According to Microsoft, the problem will be addressed in the following version, there is a Connect issue open. There is even a whitepaper, Performance Considerations for Entity Framework 5, which talks about some of the changes and optimizations coming on version 5, but by reading it, I got even more concerned:

“Once the cache contains a set number of entries (800), we start a timer that periodically (once-per-minute) sweeps the cache.”

Say what?! The next version of Entity Framework will spawn timer threads?!

When Code First came along, I thought it was a step in the right direction. Sure, it didn’t include some things that NHibernate did for quite some time – for example, different strategies for Id generation that do not rely on IDENTITY columns, which makes INSERT batching impossible, or support for enumerated types – but I thought these would come with the time. Now, enumerated types have, but so did… timer threads! Sad smile I’m afraid Entity Framework is becoming a monster.



  • The problem here is more generally with the current lack of support for filtered includes to drive the SQL generation. Without that you're left with these poorly performing outer joins and you quickly find yourself in cartesian hell. For now, we just have to carry around in our heads a list of EF-specific do's and don'ts (and the example you give above, though contrived, would unfortunately make such an EF-experienced user cringe and say "no!!!! you're doing it wrong* !" (* for some definition of 'wrong'). The good news is I know they'll get there some day (it's not like NHibernate was without its warts three releases in). Let's just hope it's not always to be fixed in the ever-elusive vNext.

  • EF is very late and a quite a bit ill-formed. From the beginning EF was quite a bit behind nHibernate and still lags substantially behind it today after several releases. MSFT claims EF is much more than an ORM but if it can't at a minimum match the ORM features of something like nHibernate than why would the community adopt it? Getting parity with NH and then convincing folks to adopt EF over NH is a very tough road to go down.

    In my opinion we would all be better served if MSFT just invested in NH and re-plumbed the EF tooling on top of it and abandoned the ORM features. Instead of throwing R&D money at something that has a low likelihood of adoption why not throw a fraction of those dollars at NH which already has a high adoption rate?

  • +1

    Don't see myself ever using EF unless I'm forced too. This is an overall distraction and waste of precious resources.

  • Scott @ Kelly:
    Thanks for your feedback. Although I agree with you, I don't see Microsoft letting go of EF... Choice is always good, IMHO, and it does integrate better with other MS technologies - see and tell me if you agree.

  • Jorge:
    NHibernate is great! If you need any help, check out NHibernate Portugal (!forum/nhpt) or drop me a line.

Comments have been disabled for this content.